Mission Control
Free & PremiumMission Control shows your Claude sessions and their agents live: on the left the sessions, grouped by project, in the middle the chosen conversation as a stream, next to it its agents as small live terminals. From here you start new sessions, write to Claude, answer its questions and switch mode, model and effort without going to the terminal. Mission Control opens as a page in the editor area.
Mission Control reads Claude Code's own transcripts and spends no tokens doing so. Claude itself always runs in one of the editor's terminals, either in a tab of the terminal window or invisibly in the background. What you type in Mission Control goes to that terminal.
How to open it
- Through the Mission Control icon in the bottom stripe, out of the box right next to the terminal.
- Menu View › Tool Windows › Mission Control.
- With Premium, Start Claude and ⌘⎋ lead here too when Settings › AI › Open Claude in is Automatic or Mission Control (Premium part).
- Through the desktop widgets (Widgets).
What you do here
- Find, rename, end and remove sessions: Sessions.
- Group sessions by project and create projects with defaults: Projects.
- Start a session with folder, access, brain, model and mode: New session.
- Read along and write messages: Conversation and Slash commands.
- See what Claude has started: Agents, Views and Background work.
- Answer questions and approvals: Questions.
- Operate Claude's screen directly: Terminal view.
- Switch mode, model, effort and Claude in Chrome: Mode, model, effort.
- Let Claude work without permission prompts: Skip permission prompts.
- Work with several accounts or API keys: Claude access.
- Give a session a Second Brain: Second Brain of a session.
- Let work go on after a usage limit or an update: Going on after a limit or an update.
- Sign Claude Code in: Sign-in.
Page layout
- Left column: New session, the project picker, the switch Skip permission prompts, the Sort and the sessions, grouped by project. Drag the right edge to widen the column; Collapse list turns it into a narrow rail with one dot per session.
- Middle: the session header with status, mode, model, effort, access, brain and context, below it the conversation stream and the message field.
- Right: the conversation's agents and background work in one of four views. Collapse agents gives the conversation more room.
When the page is narrow, the list collapses by itself and the agents move below the conversation.
Free and Premium
Mission Control is free: starting sessions, reading along, writing, answering questions, switching mode and model, Claude access and Skip permission prompts. Premium covers:
- having Start Claude and ⌘⎋ lead here instead of the terminal,
- creating and changing projects,
- a Second Brain for a session.
Sessions Free
Mission Control's left column lists your Claude sessions: the running ones and the earlier conversations of the open and recently opened projects. A click shows a session in the middle. From here you rename sessions, end them or take them off the list.
Where to find it
Open Mission Control through its icon in the bottom stripe. The list is on the left. If it is collapsed, click Expand list.
Layout of the column
From top to bottom:
- New session: opens the dialog for a new session (New session).
- The project picker: All projects or a single project (Projects).
- The switch Skip permission prompts (Skip permission prompts).
- Sort with the chosen order.
- The sessions, grouped by project. Sessions that belong to no open or recently opened project come last, under Not assigned.
If there is no session yet, it says "No Claude conversation in this project yet. Start one and it shows up here at once."
A row
| Part | Meaning |
|---|---|
| Dot | colour by status: working, ready or idle |
| Title | the conversation's name, or your own name if you gave it one |
| Shield icon | the session runs without permission prompts (bypassPermissions) |
| asks | Claude is waiting for an answer from you |
| sign in | the login has expired (Sign-in) |
| waiting until … | Mission Control goes on after a usage limit at that time (Going on after a limit or an update) |
| Clock | how long the running round has taken so far |
| Second line | for running sessions the status (working, compacting, ready, starting, waiting for sign-in), otherwise when it last ran |
| Third line | brain (name and kind), project or folder, model and Claude access, as far as known |
| Bar | for running sessions how full the context is in percent and the cost so far |
When you hover over a row, two buttons appear on the right: End (running sessions only) and the X for Remove from list or End and remove.
Sort
A click on the sort opens the choice:
| Order | Sorts by |
|---|---|
| Working now (default) | sessions with a running round first, below them by last activity |
| Start time | the latest start, newest first |
| Last active | the end of the last round, newest first |
| Active first | open sessions before ended ones, each by last activity |
Whatever you choose: Errors and questions always come first. Sessions with an error or an expired login come first, then sessions with an open question. The choice applies in every window and project.
How to do it
Rename
- Double-click the title, or right-click and choose Rename (F2).
- Type the name (at most 60 characters) and press Enter. Esc cancels.
The name applies everywhere in the editor where the session appears. An empty name brings back the original title.
End
- Click End on the row or choose it from the right-click menu.
- Confirm. If the session is working right now, the dialog warns that what it is doing is cut off and suggests Cancel.
Claude is ended, the conversation is kept and can be resumed later. If the session runs in another app, it is ended there. If several Claude processes run the same conversation, the button ends all of them and names the number, for example End 2 Claude processes.
Remove from the list
Remove from list takes an ended session off the list. For a running one the entry is called End and remove. The history stays with Claude Code.
Below the list there is then Show removed with the number. There, Back into the list brings a session back, and Hide removed closes the group again.
Resume
Pick an ended session. The conversation header shows Resume here: Claude starts in the background with this conversation. You can also just write a message; the message field then resumes the session with it.
Adjusting the list
- Width: drag the line at the right edge of the list. A double-click restores the standard width.
- Collapse list: the column becomes a narrow rail with New session, the project picker, Skip permission prompts and one dot per session. A right-click on a dot opens the same menu as on the row. Expand list brings the column back.
- When the page is narrower than about 1,180 pixels, the list collapses by itself unless you opened it on purpose.
Troubleshooting
| What you see | Why | What to do |
|---|---|---|
| A session is missing | It belongs to a project that is not picked right now, or you removed it | Pick All projects or Show removed |
| “…” could not be ended. | The Claude process does not respond | Look in the session's terminal |
| Rename is greyed out | Claude has not given the conversation an id yet | Wait a moment until the session runs |
See also
Projects Free & Premium
Mission Control groups your sessions by project. A project here is every folder you have open in the editor and every folder of the recently opened projects, each with its subprojects. On top of that come Claude projects: named projects with folders and defaults that new sessions start with, such as a fixed Claude access, a brain, a model or an effort.
Where to find it
- The project picker sits in Mission Control's left column below New session.
- Above the groups is the heading Projects with the button New project.
- Each group has a header with the project name and its own buttons.
The project picker
| Entry | Shows |
|---|---|
| All projects | every session, grouped by project |
| each open project | only that project's sessions. A subproject sits under its project with ↳, a window with several folders as a group of its own under its name |
| Recent projects | the recently opened projects that are not open in any window right now |
| Manage projects… | the list of your Claude projects |
The choice applies to this window and is kept. When you pick a project, New session starts in that project.
The groups
With All projects, every open project is a group in the order of its windows, even without a session (No session yet). Recently opened projects only appear when they have sessions. Whatever belongs to none of them comes last under Not assigned.
A group header shows:
- a purple dot when a session in it is working right now,
- the name and the number of sessions,
- + for New session in … with this project,
- the pencil Edit project when a Claude project belongs to the folder.
A click on the name or the number collapses or expands the group. When collapsed, a number shows how many sessions in it are waiting for you (a question, a sign-in or an error). The window remembers which groups are closed.
Which project a session belongs to
- If you start it through New session with a project, it belongs to that project, even if its start folder is somewhere else.
- Otherwise its folder decides: it belongs to the project whose folder holds it most closely, a subproject before its project.
How to do it
Create a Claude project
- Click New project (above the groups or in Manage projects…).
- Enter a Name, for example "Client Miller".
- Under Folders, add at least one folder with Add folder. Every session in one of these folders or below belongs to the project.
- Under Defaults for new sessions, set what new sessions start with: Claude access, Brain, Model and Effort.
- Click Save.
Edit or delete a project
- Click the pencil in the group header, or open Manage projects… and click the pencil there.
- Change name, folders or defaults and click Save.
- To delete, click Delete and then Really delete?. The sessions stay and are then listed under "No project".
Fields and options
| Field | Meaning | Default |
|---|---|---|
| Name | what the project is called | empty, required |
| Folders | full paths, at least one | the window's folder |
| Claude access | the access for new sessions (Claude access) | Default |
| Brain | a Second Brain for new sessions (Second Brain of a session) | No brain |
| Model | Opus, Fable, Sonnet, Haiku | Claude's default |
| Effort | low, medium, high, xhigh, max | Claude's default |
What happens in the background
Claude projects are the same projects as in WebDonkey's Claude usage, where you see the usage per project and per brain. They are stored in WebDonkey's database. Without WebDonkey the editor uses a database of its own. For every session the editor starts, it notes which project, brain, access, model and effort it ran with. When it is resumed, it starts with them again.
Free and Premium
The groups and the project picker are free. Creating and changing Claude projects is Premium. Deleting works without Premium too.
Troubleshooting
| Message | What to do |
|---|---|
| Creating and changing projects is part of Premium. | Activate Premium |
| Without a WebDonkey database there are no projects. | The editor has no database right now, for example because WebDonkey is older than the editor needs. Update WebDonkey |
| Please add at least one folder. | Add a folder |
| Folders must be full paths. | Pick the folder with Add folder |
| Claude does not accept this model name. | Pick a model from the list |
| This project no longer exists. | Open Manage projects… again |
See also
New session Free & Premium
The New session dialog starts Claude with everything the session needs: project, start folder, Claude access, brain, model, effort and permission mode, plus an optional first message. Claude starts in the background, no terminal opens. The session shows up in Mission Control right away.
Where to find it
- New session at the top of Mission Control's left column.
- + in a project group's header (New session in …): the dialog starts with that project.
- With the list collapsed, the + in the narrow rail.
Without a choice, the dialog takes the project shown in the project picker. With All projects it takes the project of the session currently shown, otherwise the window's project.
How to do it
- Click New session.
- Under Project, pick the project. This fills in Start folder and the project's defaults.
- Check the Start folder. You can change it; the session still belongs to the chosen project.
- Pick Claude access, Brain, Model and Effort as needed.
- Choose the Mode.
- Optionally write the first message in the bottom field.
- Press Enter in the first-message field or click Start.
Esc or Cancel closes the dialog without starting anything.
Fields and options
| Field | Meaning |
|---|---|
| Project | No project or one of the open and recently opened projects, every folder and subproject on its own. Next to it the pencil Edit project and New project (Projects) |
| Start folder | Claude starts in this folder. You can type, suggestions come from the project. The folder button opens Choose folder |
| Claude access | which account or API key Claude runs with. The pencil next to it opens Manage Claude access (Claude access) |
| Brain | No brain or a Second Brain, Premium only. Choose folder… adds any folder as a brain. The pencil opens Manage brains (Notes › Second Brain) (Second Brain of a session) |
| Model | Claude's default, Opus, Fable, Sonnet or Haiku |
| Effort | Claude's default, low, medium, high, xhigh or max |
| Mode | Default, Accept edits, Plan, Auto or Skip permission prompts. Hovering over Skip permission prompts shows a note on what the mode does |
| Claude in Chrome | whether the session starts with Claude in Chrome, Claude's connection to the browser |
| first message | "First message (optional), Enter starts" |
What the dialog remembers
- the start folder and the mode you used last,
- per project the brain you last started with there, No brain included. If someone changes the Claude project's defaults, its default applies again.
With No project the brain is always No brain. There is only a brain if you pick one, even when the start folder lies inside a brain.
The mode with a Claudian brain
If the start folder lies in a Claudian brain you picked, Claude starts the way Claudian starts it. The mode from Claudian's settings is preselected, with "Mode from Claudian's settings: …" below it. If you pick another one, it applies to this session only.
Skip permission prompts
The first time you pick Skip permission prompts, the editor warns you once and then turns on the switch in the left column (Skip permission prompts). If your Claude Code settings forbid the mode, the button is greyed out.
What happens in the background
Mission Control starts claude in an invisible editor terminal, in the start folder and with the chosen values. Model and effort go to Claude Code as --model and --effort, the mode as --permission-mode. Mission Control types the first message in as soon as Claude is ready. If the session has a brain, the message goes out the way Claudian sends it: with the note of that brain open in the editor and your selection in it. The dialog shows both as chips above the message field, and their X takes them out.
For every session the editor notes project, brain, access, model and effort. WebDonkey's Claude usage uses this to count usage per project and per brain.
Free and Premium
The dialog is free. The Brain field only appears with Premium. Creating and changing projects is Premium too.
Troubleshooting
| What you see | Why | What to do |
|---|---|---|
| Start is greyed out | The start folder is empty | Enter or choose a folder |
| A red line below the fields | Claude could not be started, for example because the folder is missing | Check folder and access |
| "No brain yet. Create one with the pencil (Notes › Second Brain)." | There is no brain yet | Create one through the pencil (Second Brain) |
| A brain shows "(not reachable)" | Its folder is not there, for example on a disconnected drive | Connect the drive or pick another brain |
See also
Conversation Free
The middle of Mission Control shows the chosen conversation as a stream: your messages (You), Claude's answers (✻), what Claude is doing right now, and edits as a small diff. At the bottom you write to Claude, the header shows and switches what the session runs with.
Where to find it
Pick a session in the left column. If none is picked, it says Pick a conversation on the left.
The header
| Element | Meaning |
|---|---|
| Title | the session's name |
| Status | working, compacting, ready, starting, waiting for sign-in, or for ended sessions done, error, idle |
| Mode, model, effort | switchable (Mode, model, effort) |
| Key | the session's Claude access (Claude access) |
| Brain | the session's Second Brain (Second Brain of a session) |
| Context | how full Claude's context is, yellow from 70 %, red from 90 % |
| Amount, ↓ | cost so far and the number of output tokens |
| Folder | the folder Claude runs in |
On the right, depending on the situation, are Resume here, Open here, Claude in Chrome, Show terminal view, Stop (ends Claude cleanly) and Show terminal.
The stream
A bar lies above the stream: Filter rows searches the stream, Expand all and Collapse all open or close all long entries together. If the filter finds nothing, it says No matching rows.
You open long entries one at a time with Show more and close them with Collapse. What Claude is doing shows as a short line: "reads …", "writes …", "runs …", "searches …", "looks up …", "starts an agent" or "thinking". Mission Control shows edits as a diff with "3 lines added" and similar.
Older messages
Mission Control first reads the end of a long conversation. The top of the stream then shows Load older messages. Each click fetches another piece before it.
Rewound
If Claude Code rewound the conversation with Rewind, the message it went back to carries the label rewound. What came after it was discarded and is no longer shown.
Errors from Claude's API
If an error from Claude's API comes instead of an answer, for example no connection or an expired login, it shows as a red row of its own, Error, with the time and Claude Code's wording. For the latest of these rows Mission Control offers to go on:
- Send again if Claude had not started working on the message yet. The message goes out once more.
- Let it go on if Claude was already in the middle of it. Mission Control sends "Please go on where you were interrupted."
More entries in the stream
| Entry | Meaning |
|---|---|
| Context compacted | Claude Code summed up the conversation to free context, with /compact or automatic, plus the tokens before and after. Show summary opens it |
| Recap | what Claude Code tells you when you come back to a conversation: where the work stands and what comes next, with the time |
| Scheduled | a job Claude created: once or recurring, possibly durable (kept when Claude Code restarts), when it runs, next run, its id (Job …) and the message it will send. If Claude Code refused it, it says not scheduled |
| Notice from Claude Code | a notice next to Claude's prompt, shown as a card. For Bug report drafted you choose Review, Send or Dismiss. It is only sent with Send |
| Final report, Message from an agent, Message from another session | messages between sessions and agents |
| Agent stopped, Shell stopped, Task stopped | something was stopped, or not stopped if that failed |
| limit reached | Claude's usage limit (Going on after a limit or an update) |
Below the last entry, Claude is working … runs with the clock during a round, and Compacting context … while compacting. If Claude is waiting for you, it says waiting for you.
Writing a message
The field at the bottom (Message to Claude, Enter sends) goes to the terminal the conversation runs in.
| Key | Effect |
|---|---|
| Enter | sends |
| ⇧Enter | new line |
| ↑ | brings back earlier messages, or edits waiting messages if there are any |
| ⌘↩ | throws the message into the running round right away |
| Esc | stops the running round (Stop the round (Esc)) |
| Tab | takes Claude's suggestion for the next message (Suggested by Claude) |
| / | shows Claude Code's commands (Slash commands) |
The plus button Attach files or folders puts files as chips above the field (Send a selection to Claude). If the session has a brain, the chips for the open note and your selection join them.
Status of a message
Until Claude has a message in its transcript, it sits as a bubble below the stream, formatted like the stream itself, with its status:
| Status | Meaning |
|---|---|
| sending … | going into the terminal right now |
| waiting, Claude is still working | follows after the running round |
| waiting for the message before it | Claude gets one message after another |
| waiting for Claude to start | Claude is still starting |
| waiting for the limit to reset | follows once the usage limit resets, with the time |
| not delivered | with Send again and Edit |
↑ brings a waiting message back into the field. It then says "Editing a waiting message · Enter puts it back · Esc cancels".
Session not running, or running elsewhere
- Ended session: the field shows "Message, Enter resumes the session in the background". Your message starts it again in the background. Resume here in the header starts it without a message.
- Claude Code background session: Open here opens it in a terminal of its own, and it keeps running. Your first message does the same.
- In another app: the field names the app and says Here you follow along. Continue here anyway starts a second Claude for this conversation. Nothing is ended in the other app. If it keeps running there, both write into the same conversation.
Width and agents
Drag the handle between conversation and agents to share the width, a double-click resets it. When the page is narrow, the agents sit below the conversation and the handle shares the height. In an agent's field you write to it through the main session (Agents).
Troubleshooting
| What you see | Why | What to do |
|---|---|---|
| not delivered | The terminal did not take the message | Send again, otherwise check the terminal view |
| "Mission Control does not recognise what Claude is showing." | Claude shows something Mission Control does not know | Show terminal view and answer there |
| There is no transcript for this conversation yet. | Claude has not written anything yet | Send the first message |
| A red Error row | Claude's API did not answer | Check the connection, then Send again or Let it go on |
See also
Slash commands Free
Type a / at the start of Mission Control's message field and Mission Control shows the commands Claude Code would offer you in its terminal: its own, your commands from ~/.claude/commands, the project's, plus skills, plugins and prompts of MCP servers. It is Claude's own list, not a copy next to it. This also works when the session runs in a tab of the terminal window.
How to do it
- Click into the message field of a running session.
- Type
/and the first letters, for example/mo. Claude Code commands appears above the field, and the list filters as you type. - Pick a command with ↑ and ↓. The picked one is marked.
- Tab or Enter takes it into the field, and so does a click on a row.
- Add arguments if needed and send the message with Enter.
Esc closes the list. The key help sits in the list header: "↑↓ choose · Tab/Enter accept · Esc close".
Notes instead of a list
| Note | Meaning |
|---|---|
| Claude Code is looking up matching commands … | Claude is building the list right now |
| Claude Code suggests commands once the conversation runs here. | The session does not run in a terminal Mission Control can reach. Resume it or start it again |
| Claude Code cannot suggest anything right now | A question is open in the terminal or text is already typed there. Answer the question or clear the prompt in the terminal view |
What happens in the background
Mission Control mirrors the command you type into Claude's prompt line in its terminal and reads back the list Claude Code shows there. It only sends letters, Backspace, arrow keys and Tab, never Enter or Esc. Text you typed into Claude's prompt yourself stays untouched. As soon as you leave the command word, Mission Control takes its mirror out of the prompt again.
See also
Agents Free
When Claude starts agents, they appear to the right of the conversation in small windows of their own, with everything they are doing. Lines connect them to the point in the conversation where Claude started them. While none runs, it says No agents in this conversation.
What an agent window shows
- Title and status: working, done, error, idle, ended (no report) or stopped.
- doing: a summary from Claude Code itself, as in the agent bar of its terminal.
- The agent's latest lines. A click enlarges the window, the icon at the top right shows everything in Full view.
- The number of calls and how full the Context is.
- Messages: what main session and agent wrote to each other, newest first (Brief from main, Report, Last message).
- If a question comes from the agent, it carries from agent "…".
Talking to an agent
In an agent's field ("Message to this agent") you write to it. The message goes to the agent through the main session, and "via" with the session's name is shown above it. If the agent is done, your message picks it up again through the main session. If the conversation is not running, it starts in the background for this.
Stopping an agent
- Click Stop on the agent.
- Confirm the question Stop agent with Confirm.
This only works while Claude is not waiting for an answer. Otherwise the editor tells you why. If the conversation runs outside the editor, stop it there with /tasks. If there was text in Claude's input box, it ends up in Claude's stash: Ctrl+S in the terminal brings it back.
Making room
- Collapse agents gives the conversation the full width, Expand agents brings the agents back. Collapsed, the rail shows how many are working.
- The view decides how the agents are arranged.
Troubleshooting
| Message | What to do |
|---|---|
| "Claude is waiting for an answer in the terminal: answer first, then stop." | Answer the open question, then stop again |
| "Several running agents have this name: stop it in the terminal with /tasks." | Use /tasks in the terminal view |
| "Stopping did not work: try /tasks in the terminal." | As above |
See also
Views Free
To the right of the conversation, Mission Control shows the session's agents and background work. The View switch at the top offers four layouts for them. The choice is kept.
The four views
| View | Shows |
|---|---|
| Tiles | a live screen excerpt for each entry |
| List | one line per entry with Kind, Title, Status, Run time and Last line. A click on a column heading sorts by it |
| Columns | grouped by status: Working, Waiting for you, Finished, Failed |
| Tree | the conversation, its agents and what they started |
How to use it
- Pick the layout under View at the top.
- Click an entry to enlarge it.
- The icon at the top right of an entry shows it in Full view. Esc or Close goes back.
The filters of the background work decide which entries appear. If nothing matches the filter, it says Nothing to show with this filter.
Tip
Columns is the quickest view to see who is waiting for you. Entries with an open question sit under Waiting for you.
See also
Background work Free
Besides agents, Mission Control shows what Claude has running in the background: Shell, Monitor, Wakeup and Workflow. The numbers on the filters count what is running right now, like Claude's own footer ("3 shells, 1 monitor").
Filters
| Element | Effect |
|---|---|
| All, Agents, Shells, Monitors | limit the overview to one kind. Hover over one to see how many are running and how many there are in total |
| Hide finished | hides what has ended. Next to it you see how many are hidden (… hidden) |
| Finished | the collapsed group of finished entries with their number. A click opens it (Show finished again) or closes it (Hide finished) |
Both are kept.
Status
Each entry is running, scheduled, done, failed, stopped or ended (no report). Shells show their exit code (exit …).
Viewing output and stopping
- Click an entry. Its output opens.
- Stop stops a shell or task. This works while Claude is not waiting for an answer.
If no output file is known yet, it says No output file known yet. If the file is gone, it says The output file is gone (temporary files were cleaned up).
See also
Questions Free
When Claude asks something or needs a permission, it shows up in Mission Control as a card at the end of the conversation. You answer there, and the editor types the answer into the session's terminal. In the list the session then carries the label asks, and in the Columns view it sits under Waiting for you.
Kinds of cards
| Card | What you do |
|---|---|
| Claude asks a question | choose one or several answers ("several possible") or write Own answer …. With several questions you go through them with Back and Review until Review your answers and submit |
| Claude asks for permission | choose yes or no, optionally write a note for Claude first ("Note for Claude (optional), then choose Yes or No"). Show diff in the editor opens the planned change |
| May Claude work in this folder? | answer the trust question for a new folder |
| Claude shows a selection | pick one of the options |
| Claude is showing a view | a Claude Code view, such as a list. You operate tabs and further keys in the terminal view |
If the question comes from an agent, it says from agent "…". Cancel (Esc) discards the card.
If an answer does not arrive in the terminal, it says "That did not arrive in the terminal. Check the terminal view." If the question is taller than the terminal, the editor says its start is not visible and offers Answer in the terminal.
Blocked by auto mode
In the Auto permission mode, Claude Code's own check decides what may run. If it refuses an action, Mission Control shows the card Blocked by auto mode: the whole action, the reason given by auto mode and, if an agent wanted it, the agent's name.
- Choose how far the rule reaches: Exactly this action or, where it fits, Every command that starts like this or Everything in this folder.
- Choose where it applies: This project only (
.claude/settings.local.jsonin the project) or All projects (~/.claude/settings.json). - This is added to the allowed rules: shows the rule before anything happens.
- Allow and continue adds it and tells the session (or, through it, the agent) that it may go on. The permission mode stays as it is. Dismiss only hides the card.
Mission Control only offers narrow rules: never a whole tool, a program on its own or an interpreter. For protected locations, nested commands or tools without a narrow rule, the card says why there is none. If the session runs in another app, the button reads Allow: the rule is added and you tell Claude there to continue. Cards older than a day are no longer offered.
Why so careful
Claude's terminal is the truth. Mission Control only sends keys when it recognises what is on the screen. If it recognises nothing, it says "Mission Control does not recognise what Claude is showing." and you answer in the terminal view.
Troubleshooting
| Message | What to do |
|---|---|
| "The settings file was just changed. Click again." | Click Allow and continue once more |
| "The settings file is not valid JSON or is built differently. Nothing was changed." | Check the file by hand |
| "This rule is too broad and is not added." | Choose a narrower reach |
See also
Terminal view Free
The Terminal view shows Claude's screen live in Mission Control. It is the fallback when something is not recognised as a card, and the place for tabs, menus and keys only Claude Code knows.
How to do it
- Click the screen icon Show terminal view in the conversation header. The view then lies over the page.
- Just start typing: "live · keys go straight to Claude".
- Close terminal view (or the same icon) hides it again.
Show terminal instead switches to the session's real terminal, in the editor's terminal window.
When you need it
- Mission Control says "The terminal view shows it and lets you answer directly."
- The mode switch with Shift+Tab was not confirmed.
- A Claude Code view has tabs or a long list ("The list goes on. You can scroll it in the terminal view.").
- A question is taller than the terminal.
The view only exists for sessions running in one of the editor's terminals. If the session runs in another app, you can only read along.
See also
Mode, model, effort Free
The conversation header holds chips that switch the running session without going to the terminal: permission mode, model and effort. Next to them you turn Claude in Chrome on or off for the session. Mode, model and effort switch without a restart.
Permission mode
A click on the chip (Permission mode · click to switch) opens the choice:
| Mode | Meaning |
|---|---|
| Default | Claude asks before commands and edits |
| Accept edits | file edits run without asking |
| Plan | Claude only plans and changes nothing |
| Auto | Claude Code's own check decides what may run |
| Skip permission prompts | only with the switch of the same name (Skip permission prompts) |
Mission Control switches like Claude's Shift+Tab, between two rounds, and then checks whether the mode arrived. The chip shows how the switch is going: queued, switching … or not confirmed. With not confirmed a note names the reason, and you switch in the terminal view with Shift+Tab. Not every model offers Auto. If Mission Control does not know the mode yet, it says mode unknown.
If the session is not running, the chosen mode applies from the next resume.
Model
The Model chip switches the running session with /model. Claude Code also keeps the pick as the default for new sessions. Open Claude's own selection… shows every model, the effort and "this session only" (key s).
The switch does not work while Claude expects an answer, a started text sits in Claude's input line, Claude is still starting or a message is being typed. The reason is shown. If Claude is working, the new model applies from its next step.
Effort
The Effort chip switches the running session with /effort, nothing restarts. If Claude is working, the new effort applies from its next step.
Claude in Chrome
The Claude in Chrome switch applies per session. Hover over it for its state: On in this session., Off in this session. or "Requested but not active" (then /chrome in the session shows why).
- A click switches it. Claude restarts briefly for this, the conversation continues. If Claude is working, it happens once Claude is done.
- Right-click, then Choose browser…, sets which Chrome is paired.
- For an ended session the value applies from the next resume.
Editor reference
The ⧉ switch at the message field decides whether this conversation gets the open file and selected lines from the editor (Send a selection to Claude).
See also
Skip permission prompts Free
The switch Skip permission prompts lets Claude run all commands and file changes without asking. Mission Control uses Claude Code's mode bypassPermissions for this, the same one Claude's VS Code extension offers. Nothing of Claude Code itself is changed or worked around.
Where to find it
The switch sits at the top of Mission Control's left column, below the project picker. With the list collapsed it is the shield icon in the narrow rail. The New session dialog offers the same mode as a button under Mode.
How to do it
- Click Skip permission prompts.
- The first time, the editor warns you: Claude will then run everything without asking first, including deleting, installing and network access. Use it only in projects you trust, ideally under Git or with a backup. You yourself are responsible for everything Claude carries out in this mode.
- Confirm with Turn on. The warning appears only once.
You can turn it off at any time with one click, without a confirmation. Claude then asks for permission again.
What the switch does
| State | Effect |
|---|---|
| Off (default) | Claude asks before commands and file changes, in the mode you chose. No session starts without permission prompts |
| On | New and restarted sessions run in the mode bypassPermissions. Running ones switch to it where Claude Code allows that without a restart |
The switch applies to every window and project.
Running sessions
- A session Mission Control started with the switch on can move between modes without a restart.
- A session started without the switch only gets the mode with its restart. Its mode chip then shows after restart. Hover over it for a note that the switch applies to this session once it restarts.
- When you turn the switch off, running sessions in this mode go back to Default.
Sessions running without permission prompts carry a shield icon in the list ("Runs without permission prompts (bypassPermissions)"). The mode chip in the header shows Skip permission prompts.
What happens in the background
Mission Control starts Claude with --permission-mode bypassPermissions and --allow-dangerously-skip-permissions. The second flag keeps the mode in Claude's Shift+Tab cycle, so you can switch out of it and back in without a restart. Mission Control writes nothing into your settings for this.
Troubleshooting
The switch is greyed out when Claude Code settings forbid the mode (disableBypassPermissionsMode):
| Note | Source |
|---|---|
| "Not available: your organisation's Claude Code settings forbid this mode (disableBypassPermissionsMode)." | a managed setting of your organisation |
| "Not available: your Claude Code settings forbid this mode (disableBypassPermissionsMode in settings.json)." | your own settings.json |
Then no session starts in this mode either.
See also
Claude access Free
With Claude access you work with several Claude accounts or API keys side by side in the editor, say one for your own projects and one for a client. Every Claude session the editor starts runs with exactly one access. One of them is the default, and a Claude project can set its own.
Where to find it
- In the New session dialog, next to Claude access, the pencil Manage Claude access.
- In a project's form under Defaults for new sessions, the same pencil.
The dialog is called Claude access.
Kinds of access
| Access | What it is |
|---|---|
| Claude sign-in (/login) | Claude Code's own sign-in on this Mac. Always there, cannot be removed |
| Claude account | an account token created by claude setup-token. It is valid for a year and starts with sk-ant-oat |
| API key | a key from the Claude Console (platform.claude.com). It starts with sk-ant-api |
How to do it
Add a Claude account
- Open Claude access.
- Under Add access, enter a Name, for example "Client Miller".
- Set Type to Claude account.
- Click Create token in the terminal. The dialog closes and
claude setup-tokenruns in the terminal. It opens the sign-in in your browser. Pick the account you want there. - Copy the token from the terminal, open Claude access again and paste it under Account token.
- Click Save.
Add an API key
- Enter a Name and set Type to API key.
- Paste the key under API key. If it starts with
sk-ant-apiorsk-ant-oat, the dialog sets the matching type by itself. - Click Save. On the first start, Claude asks once whether to use the key.
Set the default
The circle in front of an access (Use as default) makes it the default. It then carries the label Default. Sessions without their own choice and without a project default run with it.
Edit and remove
- The pencil Edit loads the access into the form. If you leave the secret empty, the stored one stays ("Leave empty to keep the stored one"). Without a new secret, the type stays as it is too.
- The bin Remove asks Really remove?. A second click removes the access from the list, from the keychain, as the default and from every project.
Which access applies
- The access you pick in New session.
- When resuming, the access the session ran with before.
- Otherwise the default of the Claude project the folder belongs to.
- Otherwise the Default.
Running sessions keep their access, even if you change something here. In Mission Control a chip with a key in the conversation header shows the session's access, and in the list it appears in the line below the title.
What happens in the background
- Keys and tokens are kept only in the macOS keychain. The editor itself only holds name, type and the last four characters.
- The editor hands the key only to the one terminal the session starts in and blanks every other way Claude Code could sign in for that session. So an
export ANTHROPIC_API_KEYin your shell or anenventry insettings.jsoncannot take the session over. - Nothing global changes: Claude's own sign-in,
~/.claudeandsettings.jsonstay untouched. - As long as you have not created an access, nothing about starting Claude changes at all.
Free and Premium
Claude access is free. Saving an access as a Claude project's default is part of creating and changing projects and therefore Premium.
Troubleshooting
| Message | What to do |
|---|---|
| "~/.claude/settings.json sets … This can override access chosen in the editor." | Remove apiKeyHelper or the key in the env block there if the accesses should apply |
| This does not match the selected type: API keys start with sk-ant-api, account tokens with sk-ant-oat. | Pick the matching type |
| Please paste the key or token, without spaces. | Paste the secret without spaces |
| The keychain denied access. | Allow access in the macOS prompt |
| Claude does not start, saying the keychain entry cannot be read | Edit the access and paste the secret again |
See also
Second Brain of a session Premium
A Claude session can get a Second Brain: a folder of Markdown notes that serves as Claude's long-term memory. Claude then knows what the brain is for, fetches single notes from it as needed and writes new ones into it. The notes stay a plain folder, compatible with Obsidian, but you do not need Obsidian for this.
Where to find it
- The Brain field in the New session dialog.
- As a Claude project's default in the project form (Projects).
- On the Second Brain page of the Notes tool: Start Claude on a brain (Second Brain).
How to do it
- Click New session.
- Pick a brain from the list under Brain. Choose folder… adds any folder to the list as a brain.
- Pick the start folder and click Start.
If there is no brain yet, a note appears below the field. The pencil next to it (Manage brains (Notes › Second Brain)) opens the Notes tool on the Second Brain page, where you create brains.
A project remembers the brain you last started with there and picks it again next time. No brain starts without a brain, even if the start folder lies inside one.
How to tell a session has a brain
- In the list, a brain mark with name and kind (Claudian, WebDonkey or Personal) sits below the session's title.
- The conversation header has a chip with a brain icon. Hover over it to see the path ("Second Brain of this session: …").
- A session not started in Mission Control gets the mark of the brain whose folder it runs in, for example a chat from Claudian in Obsidian.
In the message field
If the session has a brain, the message field works like Claudian's in Obsidian:
| Element | Meaning |
|---|---|
| Chip Note: … | the brain's note open in the editor. It goes to Claude with the first message. A click opens it, the X (Do not send the note) takes it out |
| Chip Selection in … | selected lines in a note of the brain. They go along with every message while they are selected. The X (Do not send the selection) takes them out |
| @ | suggests notes, files and folders of the brain (Notes and folders in …), aliases included. ↑↓ choose, Tab or Enter inserts, Esc closes |
Below your sent message the stream shows what it took along, for example "Selection in … · lines …". [[Wikilinks]] in Claude's answers are clickable and open the note in the brain. If the note does not exist yet, the click creates it where the vault puts new notes.
Claudian brains
A Claudian brain is a vault you use in Obsidian with the Claudian plugin. If you start a session in a Claudian brain's folder and picked it as the brain, Claude starts exactly the way Claudian starts it: with its system prompt, model, effort, permission mode, environment, MCP servers and skills, and without the trust question at the first start. A new chat also shows up in Claudian's history with its first message.
Anything you pick differently in New session applies to this session only: a model, an effort or another mode. The dialog then says "For this session only, instead of Claudian's setting (…)."
What happens in the background
- Started in the brain itself (the start folder is the brain's folder): Claude starts the way that suits the brain's kind. A WebDonkey or personal brain gets the brain MCP server and the working guide for its kind.
- Started in another folder: the brain's folder becomes an extra working directory of the session. Claude loads the brain's
CLAUDE.mdand its skills and commands, gets a short description of the brain and of your other brains (name and path only) and the MCP serverwd-brain. The brain's.claude/settings.jsondoes not apply here. - Through
wd-brainClaude fetches single notes instead of loading the whole brain. Your other MCP servers stay as they are. - The editor writes nothing for this into the brain, the project,
.obsidian,~/.claudeorsettings.json.
What Claude reads from the brain goes to Anthropic like everything else Claude reads.
Free and Premium
A session's Second Brain is Premium, like the notes. Without Premium the Brain field is missing in New session, and a session starts without a brain.
Troubleshooting
| What you see | Why | What to do |
|---|---|---|
| A brain is shown as "Vault no longer in the list" | It was taken off the list of brains | Pick another brain |
| A brain shows "(not reachable)" | Its folder is not there | Connect the drive or bring the folder back |
| The chip Note: … is missing | The open note does not belong to this brain | Open the note from the brain |
See also
Going on after a limit or an update Free
Two things interrupt a Claude session without you doing anything wrong: your usage limit is reached, or the editor restarts for an update. In both cases Mission Control makes sure the work goes on afterwards.
After a usage limit
When a session hits Claude's usage limit, Claude Code answers with an error. Mission Control shows this in the stream and on your messages.
When Claude Code goes on by itself
For a limit that resets within a day, Claude Code usually waits by itself. Your message then carries waiting for the limit to reset with the time. Claude Code has the message and goes on with it by itself once the limit resets. Further messages you write in the meantime reach Claude after the reset as well.
When Claude Code does not go on
With the weekly limit, a reset more than a day away, or when you switched off waiting in Claude Code or cancelled it with Esc, Claude Code does not go on by itself. Your message then carries limit reached. Messages Mission Control still holds are typed in after the reset. On top of that, Mission Control takes care of going on:
- It keeps a "continue" for the session. The message tells Claude the limit has reset and that it should pick up where it was interrupted, without repeating what is already done.
- In the list the session shows waiting until … with the time, and the stream shows the same on the waiting message.
- Shortly after the reset, Mission Control types the "continue" in. If Claude Code still shows its limit dialog, Mission Control first closes it with the option that only waits for the reset.
Don't continue on the waiting message drops the "continue". It is also dropped if you write something yourself after the limit, or if Claude goes on by itself.
Mission Control takes the time from Claude Code's message, otherwise from the statusline's limits. The editor has to be running for this.
After an editor update
When you install an editor update, the editor restarts. Claude sessions running at that moment go on afterwards:
- Right before the restart, the editor writes down which Claude sessions are running: conversation, folder, project, Claude access, brain, whether it ran in a terminal or in the background, and whether a round was running.
- After the restart, every project window resumes its sessions in its terminals, in the permission mode the conversation last had. Mission Control's background sessions go on in the background right away.
- A session whose round the restart cut off gets the message "continue" once.
- Sessions with a brain go on with their brain. If WebDonkey is not yet at the level the editor needs, such a session waits up to ten minutes for it and then goes on without its brain.
If a session's project does not come back in any window, the editor resumes it in the background after a while, where Mission Control reaches it. If you quit the editor normally, it writes nothing down and nothing starts by itself.
Free and Premium
Both are free. A resumed session only gets a brain with Premium.
Troubleshooting
- Nothing happens after the limit: check whether the session shows waiting until …. If not, write to Claude yourself after the reset.
- A session is missing after the update: resume it in Mission Control with Resume here.
See also
Sign-in Free
If Claude Code is not signed in, or no longer, in a session, Mission Control guides you through signing in. You do not need to go to the terminal.
When it appears
| Card | Meaning |
|---|---|
| Login expired | A message did not get through because the login expired. In the list the session shows sign in, on the message "not answered, login expired" or "interrupted midway, login expired" |
| Signed out | Claude Code was signed out with /logout and has ended. Sign in now starts the session again |
| Not signed in | Claude Code is not signed in. Sign in now guides you through it |
| Set up Claude Code | Claude Code is not set up here yet. You pick the terminal colours first, signing in comes next |
How to do it
- Click Sign in now.
- Under Sign in with what?, choose how you sign in.
- Mission Control opens the browser ("Claude is opening the browser …"). If it does not open, use Open in browser or Copy link.
- If the sign-in page asks for a code, enter it under Code from the sign-in page (if asked) and click Send code.
- Finally Claude Code shows some notes (Notes from Claude Code). Continue confirms them.
It then says Signed in again. The login applies to every session.
Going on after signing in
- If you already signed in from another session, Send again or Send the last message again is all it takes here.
- Let it go on has Claude continue where it was interrupted.
Sessions in other apps
If the session runs in another app, sign in there with /login.
Several accounts
To start sessions with different accounts or with an API key, create Claude access entries. Signing in here concerns Claude's own sign-in on this Mac.