Linear MCP: What It Does, How to Connect It, and Where Its Limits Are
Linear's official MCP server lets Claude, Cursor and Codex file and update issues. Setup, read-only mode and limits.

You're deep in Claude Code, the feature is done, and it needs a ticket. Linear MCP is Linear's official MCP server, and it lets the agent file that ticket, or move it to In Review, without you leaving the terminal.
It's remote, so there's nothing to install: add one URL to your AI client and sign in. Below: what it does, setup in each client, the read-only mode, and the limits worth knowing before you hand an agent your board.
What is Linear MCP?
Linear MCP is the official Linear MCP server: Linear builds and hosts it, so it isn't a community project. Linear launched it on May 1, 2025, thanking Cloudflare and Anthropic for help getting it running (Linear changelog).
MCP stands for Model Context Protocol, the open standard AI tools use to call outside apps: the app exposes tools, and your AI client calls them when you ask in plain language. Linear's server sits on top of its own API, so once you authorize a client, it acts in your Linear workspace with your permissions.
The details, from Linear's MCP docs in October 2026:
- Endpoint:
https://mcp.linear.app/mcpover Streamable HTTP. The oldhttps://mcp.linear.app/sseendpoint is a deprecated fallback. - Auth: OAuth 2.1 with dynamic client registration, or a Linear API key or OAuth token sent as
Authorization: Bearer <token>. - Enterprise sign-in: workspaces on Okta can turn on "MCP enterprise managed authentication", so Claude signs users in under Okta policies.
There are also community Linear MCP servers on GitHub, many built before the official one existed. The hosted server is the one Linear maintains, so start there.
What can the Linear MCP server do?
It finds, creates and updates most of the project management work you'd do by hand in Linear. Split by access:
| Read | Write |
|---|---|
| Search and read issues, issue details and comments | Create a new issue, update issues (description, status and other fields) |
| Read projects and project updates | Create and update projects, add comments |
| Follow milestones, cycles, sub-issues and issue relations | Create and edit project milestones and project updates |
| Read initiatives | Create and edit initiatives and initiative updates, manage project labels |
The planning tools are newer: on February 5, 2026, Linear added initiatives, project milestones, project updates and project labels to the server, plus support for loading images (Linear changelog). Linear doesn't publish a full tool list in its docs, so check /mcp or your client's tool panel for the current set.
AI agents that write code get the most out of it. Claude Code in the terminal, Cursor in the editor or Codex on a task can read the issue to understand the ask, write the change, then move the ticket and file the follow-up bug it found. "Create a ticket from chat" is a party trick; "the agent that shipped the fix moves the issue to In Review and comments the PR link" is a workflow.
How do you add Linear MCP to Claude, Cursor, Codex and ChatGPT?
You add the remote URL to your client and complete the OAuth sign-in in the browser, with no local configuration in most clients; the tool then inherits your Linear permissions. Steps below follow Linear's docs unless noted.
Claude (claude.ai and Claude Desktop)
- On Free or Pro, open Settings > Connectors in Claude and add the Linear connector.
- On Team or Enterprise, connect Linear from the connectors page at claude.ai.
- Sign in to Linear and approve access.
Claude Code
- Run
claude mcp add --transport http linear-server https://mcp.linear.app/mcp. - Start a session and run
/mcpto finish the OAuth sign-in.
Cursor
- Open Cursor's MCP tools page and search for Linear, or use the install link in Linear's docs.
- Alternatively, add
"linear": { "url": "https://mcp.linear.app/mcp" }undermcpServersin~/.cursor/mcp.json(Cursor docs). - Complete the sign-in when Cursor prompts you.
Codex
- Run
codex mcp add linear --url https://mcp.linear.app/mcp. - Run
codex mcp login linearto authenticate.
ChatGPT
ChatGPT has its own Linear app rather than a Linear-documented MCP setup. OpenAI's help center says it reads issues and comments and can create tickets, and it needs a Linear workspace with API access enabled (OpenAI Help Center). On plans with developer mode, you can also add the server URL as a custom MCP server; see how ChatGPT uses MCP.
VS Code, Windsurf and Zed
Linear's docs route these through npx -y mcp-remote https://mcp.linear.app/mcp as a local command, which bridges clients without native remote support. If you've never wired up a server, here's how connecting an LLM with MCP works.
Does Linear MCP have a read-only mode?
Yes: point your client at https://mcp.linear.app/mcp/readonly, which "only ever exposes read tools", according to Linear. Two other ways get the same result: request only the read OAuth scope on the standard endpoint, or use an API key created with only the Read permission.
Use read-only for research sessions, standup summaries and planning chats where nobody needs the agent editing tickets. Switch to the full endpoint only in the sessions that should write.
Where are Linear MCP's limits?
The limits that bite in practice are write access, response size, rate limits and multiple workspaces.
Write access. An agent acting on your authorization can file duplicates, edit the wrong ticket or reassign work as confidently as it does anything else. Keep a human approving writes on a shared workspace, and default to the read-only endpoint.
Response size. Clients cap what one tool call can return. Claude Code warns when MCP output passes 10,000 tokens and cuts it at 25,000 by default, adjustable with MAX_MCP_OUTPUT_TOKENS (Claude Code docs). Asking for every issue in a busy project can hit that ceiling, so query narrowly: a specific Linear team, one status, one cycle.
Rate limits. Linear's MCP docs don't publish limits of their own. Its public API allows 5,000 requests an hour per user for OAuth apps and 2,500 for API keys (Linear developer docs); whether MCP calls draw on the same quota isn't documented.
Multiple workspaces. Reconnecting doesn't switch workspaces. Linear's docs suggest a separate mcp-remote config directory per workspace, each signed in on its own. If a connection fails with an internal server error, the same docs say to clear saved credentials with rm -rf ~/.mcp-auth and reconnect.
Is Linear MCP safe to connect?
It's as safe as the access you grant, with one real risk: ticket text is input the model reads, and anyone who can write a ticket or comment can try to steer your agent through it. That's prompt injection, and it matters most when the same session can also write somewhere else.
Keep research on /mcp/readonly, give automations a restricted API key instead of your full account, approve write actions one by one, and don't pair Linear in one session with a server that reads untrusted public content. The full threat model is in Is MCP Safe?
Should you use Linear MCP as your AI's source of context?
No. Linear MCP is built to act on Linear, not to be the standing context your AI works from. Rebuilding who you are, what your team standardized on and what was decided last sprint by querying tickets live means a tool call, and its tokens, on every turn, with answers capped by response limits.
That's the line between a connector, MCP and a context layer: a connector gives the agent a place to look and act; context is what it should already know when the session starts.
Unabyss is that context layer. It's one MCP connection that builds your personal context from Linear, GitHub, Slack, Gmail, Notion and the other apps on its integrations page, keeps it current, and serves it to Claude, ChatGPT, Cursor and Codex. It doesn't file, edit or comment on Linear issues, so many people run both: Linear MCP for the writes, Unabyss so the agent knows whose tickets they are. In an August 2026 Unabyss benchmark against connected tools alone, that context cut cost by 36%, time by 38% and follow-up prompts by 60% (What Structured AI Context Is Worth). Other vendor servers are compared in Best MCP Servers for Work.