Connect your AI and your computers to Context Board

Your AI can work on your board in two places. In its own app, where it reads the board and builds widgets over MCP. And in the chat inside the board, which runs on the Claude Code or Codex you already pay for.

This page walks through each step on the board’s Connect your tools checklist, at the bottom left.

Pick your route

You use	Do this	Time
Claude Code or Codex	Link your computer. One line gives them your boards and runs the board’s chat on your plan.	About a minute
Claude, ChatGPT, Cursor, VS Code or another app	Add the board’s address to the app and click Allow.	Two or three steps
Neither, or you’d rather pay per use	Chat on your own API key.	Paste one key

You can do all three. They work side by side.

Link your computer

The quickest way in if you use Claude Code or Codex. One line, pasted once, does three things.

• Your agent gets your boards. It adds Context Board to the Claude Code and Codex on that computer, so any new session there can read and build your board. There’s nothing to sign in to.
• The board’s chat runs on your plan. The chat, morning briefs and agent jobs run on your own Claude Code or Codex. No API key, no extra bill.
• Your sessions show on the board. It feeds the AI usage, Agent activity, Needs you and Git activity widgets.

The line installs a small program called the runner. It starts with the computer from then on and updates itself.

The steps

1. On the board, open Connect (the plug icon at the top right), then Your AI, then Your computers.
2. Pick Windows, or Mac or Linux.
3. On the computer where you use Claude Code or Codex, open PowerShell or the Command Prompt on Windows, or Terminal on a Mac or Linux.
4. Copy the line from Connect, paste it there and press Enter.

The line looks like one of these, with your own code in place of CODE.

# Windows
powershell -c "irm https://board.contextbolt.com/install.ps1?c=CODE | iex"

# Mac or Linux
curl -fsSL https://board.contextbolt.com/install.sh | sh -s CODE

The window says what it’s doing as it goes, and ends with Done. Connect shows the computer as linked within a few seconds. You can close the window.

5. Open a new Claude Code or Codex session and ask it about your board.

“Read my Context Board and tell me what needs my attention today.”

Good to know

• It brings Node.js if the computer has none.
• Each line works once. Copy a fresh one for each computer.
• Installed Claude Code or Codex later? The runner notices and adds the board to it on its next look, without another line.
• Neither signed in yet? Connect says so. Install one, sign in, and it shows up within a minute.
• Already had the board in Claude Code or Codex? The old entry is replaced, never doubled.

In Settings, under Keys, the line’s access shows as Claude Code on and the computer’s name, and Codex on and the computer’s name.

Connect your AI app over MCP

For Claude, ChatGPT, Cursor, VS Code, Windsurf and any app that can add a remote MCP server. The board’s address is the same for every app.

https://board.contextbolt.com/mcp

Your app signs in with an Allow page, so there’s no key to copy. The page names the app, says whether it’s verified, and lists what it can do. Click Allow.

Open Connect, then Your AI, and pick your app. Its page has these steps with the address filled in, and turns green the first time the app reads the board.

Claude (web and desktop)

1. Copy the board’s address from the Claude page in Connect.
2. In Claude, open Customize, then Connectors, and choose Add custom connector.
3. Name it Context Board, paste the address and add it. Choose Connect, then Allow on the board’s page.

On the Free plan you can add one custom connector. On a Team or Enterprise plan, an owner adds it once in Organization settings, then each member connects with their own account.

ChatGPT

1. Copy the board’s address from the ChatGPT page in Connect.
2. In ChatGPT, open Settings, then Apps and Connectors, then Advanced settings, and turn on Developer mode.
3. Choose Create, name it Context Board, paste the address, keep OAuth and create it. Then Allow.

Custom connectors depend on your ChatGPT plan. If a menu name doesn’t match, the page in Connect has the current steps.

Cursor

1. On the Cursor page in Connect, choose Add to Cursor. Cursor opens and asks you to install it.
2. When Cursor shows Needs login beside it, click that and Allow.

Or add it to ~/.cursor/mcp.json yourself.

{ "mcpServers": { "context-board": { "url": "https://board.contextbolt.com/mcp" } } }

VS Code

1. On the VS Code page in Connect, choose Add to VS Code. VS Code opens and asks you to install it.
2. When VS Code asks to sign in, choose Allow. Then use it from Copilot Chat in agent mode.

Or add it to .vscode/mcp.json yourself.

{ "servers": { "context-board": { "type": "http", "url": "https://board.contextbolt.com/mcp" } } }

Claude Code or Codex, without linking the computer

Linking your computer is the easier route. To add only the app, use Other ways on its page in Connect.

Claude Code: run these two lines. The second opens your browser, where you click Allow.

claude mcp add --transport http --scope user context-board https://board.contextbolt.com/mcp
claude mcp login context-board

Codex: run this line, then click Allow in the browser. If it didn’t open, run codex mcp login context-board.

codex mcp add context-board --url https://board.contextbolt.com/mcp

Start a new session afterwards. Both apps load their MCP servers when a session starts.

Windsurf and other MCP apps

Paste the address where your app adds a remote MCP server, and click Allow if the board’s page opens. In Windsurf, add it to mcp_config.json as "serverUrl": "https://board.contextbolt.com/mcp".

Apps that can’t sign in

Under Other ways on an app’s page, Make a link with a key gives you the board’s address with a key in it. It works without signing in, so treat it like a password. Remove it any time in Connected.

Project folders

Give your agent folders to work in, like your notes or your code. The board’s chat, briefs and jobs can then read them, or change them.

You need a linked computer first.

1. Open Connect, then Your AI, then Project folders.
2. Choose Add a read and write folder or Add a read-only folder.
3. Pick the folder in the window that opens on your computer.

• Read only: the agent reads the files and changes nothing.
• Read and write: the agent can change files as it works. Each file it changes or deletes is kept as it was, on your computer, for 30 days.

On another device? The picker opens on the linked computer, so the board says you’re not there. Choose Open it on and the computer’s name anyway, or use the line under Or allow a folder with a line, run on that computer.

Take a folder back with the X beside it, here or in Settings under Your computer.

Use your existing MCPs

The board’s chat can use the MCP servers your Claude Code or Codex already has, like your email or calendar. They run and sign in on your computer, never on the board.

It’s off until you turn it on, separately for Claude Code and for Codex. You need a linked computer.

1. Turn on Use my existing MCPs. It’s on the Claude Code or Codex page in Connect, under The board’s chat, and in the chat’s model menu, under Your MCPs.
2. Optional: pick a Project. The chat then also uses that project’s own settings and servers. Pick only a project you trust. None uses only the computer’s own servers.
3. Ask the chat something that needs one of them.

You approve each new call. The first time a server’s tool runs, the chat asks, for example “Let Gmail run search_messages?”. Choose Allow or Don’t. A question nobody answers in 10 minutes counts as a no. The question shows in every tab you have open.

If something’s wrong, the chat says so in one line.

• A server needs sign-in: for Claude Code, open it on that computer, type /mcp and sign in. For Codex, run codex mcp login and the server’s name. Then ask again.
• A server won’t start: its tools aren’t there this time. Check it with claude mcp list or codex mcp list on that computer.

The chat in the board

The chat opens beside the board. Use the button at the top right, which reads Ask Claude or Ask Codex once it’s set up, the command bar, or Ctrl J (⌘ J on a Mac). It builds widgets, boards and themes from a description, with the same tools as an outside agent.

It runs one of two ways. With both set up, your linked computer wins.

• On your computer: with your computer linked, it runs on your Claude Code or Codex.
• On your own API key: see below.

Every change it makes is in plain sight.

• Each tool call shows as a line that jumps to its widget.
• Undo this change, or Undo these 3 changes after several, puts widgets and boards back as they were. Data and sources it connected stay.
• A source that needs your key opens the board’s own form. Keys never pass through the chat.

Questions

When the agent can’t tell what you mean, it asks in the chat: “Which to-do list do you mean?”, with the answers to click and a box for your own. It can ask a few things at once, and take more than one answer where that fits. The reply waits until you answer, and Stop still works. On your computer, Claude Code and Codex ask with their own question tools. On an API key, the chat asks the same way. The question and your answer stay in the chat.

Chats

The Chats list sits beside the chat: every chat you’ve had, with Claude and with Codex, in one list. Chats still going come first: one working on a reply, one waiting on your answer, or one with a reply you haven’t read yet. Then the ones you pinned, then the rest by day. Find one with the search box, and rename, pin or delete it from its row.

Open any chat and it carries on in the same session. A reply on your computer keeps going in its own chat while you open another, start a new one or switch between Claude and Codex, and two chats can reply at once. A reply on an API key runs in the page, so the chat waits for it.

Your chats are kept on your board, so they’re on every device you sign in on. The board keeps your newest 200, and the ones you pinned however old. Fold the list away with the list button in the chat’s header; it stays folded until you open it again.

Chat on your own API key

No Claude Code or Codex? Choose Or use an Anthropic API key instead when you set up the chat, or use an OpenAI key for Codex, then Start. The key stays in your browser and goes only to that provider, who bills you for what the chat uses. On this route, the chat asks before it removes anything or reaches a new address. Remove the key from the chat’s model menu, with Remove the key from this browser.

What your agent can do on the board

Plain asks work. “Add this to my to-do list” finds your list. Every tool takes a widget’s title as well as its id, so the agent finds “the clock” or “my orders goal” the way you name it. When two things fit, it asks which. In the board’s chat, it asks with answers to click.

• Read: the whole board, any widget, and the apps and services you connected.
• Build: add, change, arrange and remove widgets, and send them data. When no built-in widget fits, it writes a custom one.
• Work on board types: write doc pages, fill in a sheet, write a slide deck, plan a build and keep the plan current, grow a mind map, put tasks on your Calendar and plan your day, draft posts on a content calendar and log deals on a client pipeline. See Board types.
• Connect: add free and public sources the board refreshes itself, and show data from services you connected.
• Keep you posted: add and tick tasks, write notes, post an update on what it found, and say what it’s working on.
• Make it yours: set or make a theme, and make or rename a board.
• Show you: when you ask (“open it”, “show me my calendar”), it opens a board in your tab, one it just made too, or goes to a widget, a doc page, full screen or All boards. It never moves your view on its own. With no board tab open, it says so.

An agent deletes a board only when you ask it to, and the board goes to the trash for 30 days, where you can bring it back. It can’t share a board, and it never needs your keys. You connect a service once, in Connect, and the agent reads through that connection.

A few widgets are yours to add, because they read your own computer or a product you sign in to. AI usage, Needs you, Git activity, Agent job and Project file come from the gallery (press A). The Agents template already has the first three.

Everything an agent does shows at once, with its name on it, and can be undone. Deleted widgets go to the trash for 30 days.

In Claude Code, the board’s make_widget and make_theme prompts work as slash commands.

Briefs and agent jobs

Both run on your linked computer, with your own Claude Code or Codex.

• Morning brief: turn on Write it every morning in the brief’s settings and pick a time. Your agent writes the one thing that matters today, what changed, what needs you and the three things to do first. Write it now runs it at once.
• Agent job: add an Agent job widget, or choose Give it a job in a Number, Chart, List or Table widget’s menu. Say what it does in your own words, and choose Every day, Every week or When I run it. Run now runs it at once, and the answer lands on the widget.

Only you give a widget its job or set a brief’s schedule. An agent can write a brief, never set one to run.

More than one computer, or a server

Link each one with its own line from Your computers. Sessions, usage and git come from all of them.

With two or more, one is the main one. It runs your board’s chat, briefs and jobs, and its project folders. Choose Make it the main one to change it. While the main one is off, another that is on takes over.

What leaves your computer

From your sessions, only names, states and percentages. That means a session’s title and project, whether it’s working, done or waiting for you, and how much of each limit is left. Never what was said in a session.

To offer your MCP servers on the board, the runner also sends each server’s name and web address. Never a key, token or command. A server whose address may hold a key, or that runs as a program on your computer, is sent by name only.

For the Connections widget, it sends each server of Claude Code’s and Codex’s with its state, where it’s set (like ~/.claude.json, or the project folder), its domain or the program it runs, and its tools’ names. Never a key, token, password, header or the rest of an address. The board strips anything that looks like a secret a second time when it arrives.

Two widgets send a little more, and only if you add them. Git activity sends each repo’s name and branch, how many files have changes, and recent commits (their short hash, author and message), never your code. Project file sends the text of the one file you point it at, so it can show it.

A shared view of your board never shows your sessions, usage, jobs or brief settings. To hide your sessions, turn off Show my Claude Code and Codex sessions in Settings, under Your computer.

Unlink a computer

Choose the unlink button beside it in Connect, Your AI, Your computers, in Connected under Computers, or in Settings.

Its sessions leave your board and the runner there stops. The Claude Code and Codex on that computer lose their access to your boards too. Link it again any time with a fresh line.

The AI usage widget

AI usage shows what’s left of each plan’s current five-hour window and its week, for Claude Code and Codex, as rings that empty as you use them. A ring is green at half or more, amber from a fifth, and red under that. Each says when it resets, like “resets in 1h 42m” or “resets Mon 9:00 AM”, and a model’s own weekly limit shows when your plan has one.

Set an Alert in its settings: Off, or 10, 20, 30 or 50% left, with 20% to start. Below it, the ring turns red and the board shows a notice, once a window.

It reads what each app last heard from its own servers, so it fills after the app’s next use. Add it from the gallery, or start from the Agents template. More on the Claude Code usage and Codex usage pages.

The Connections widget

Connections counts every MCP server and connection your agents and your board use, and says how many need you. It’s red while one is failing, amber while one waits for a sign-in, and calm otherwise. Click it for the full view.

It reads Claude Code and Codex on each linked computer. That covers your own servers, claude.ai connectors, plugins, a project’s .mcp.json, and the apps on the ChatGPT account Codex uses. The board’s own apps, services and keys are in it too. It doesn’t read Cursor, VS Code or Claude Desktop.

The full view starts with each computer and the board, and how their connections are doing. Below that is what needs you.

• Failing: the server wouldn’t start, or turned down its key. The reason is beside it.
• Sign in: it’s waiting for you to sign in to it.
• Doubled up: one agent reaches the same service two ways, like GitHub as a claude.ai connector and as your own server.
• Unused: a key or app of the board’s that nothing has used for 30 days.

Then every server, grouped by where it comes from, with a search box. Each one shows which of its tools can change things and which only read.

It never changes an agent’s settings. For a server of Claude Code’s or Codex’s, it shows the command to run on that computer, like claude mcp remove "old-db" -s user, with a button to copy it. Run it, then press Look again. A claude.ai connector or a ChatGPT app gets a link to where it’s managed instead. The board’s own connections you can sign in to again, remove or revoke right there.

It never starts a project’s servers. A .mcp.json can run programs on your computer, so the runner reads project servers from the file and marks them Not checked, or Not approved when Claude Code hasn’t been told to use them. Your own servers and connectors are checked the way a new session checks them, from your home folder.

It looks every six hours while the computer is on and the widget is on one of your boards, and when you press Look again. A computer that’s off keeps what it last read, and says when. Connections is yours alone. On a shared view, or for a coworker, it says Private and shows nothing.

Add it from the gallery, or ask your agent to add it. It needs runner 0.23.0 or later, which updates itself. More on the manage MCP servers page.