Linear MCP server

What it does

Linear hosts this server at https://mcp.linear.app/mcp. It covers far more than issues: an agent can also run projects and milestones, write documents and status updates, manage labels and releases, and review and merge pull requests through Linear's diffs. Every call acts as the Linear user who signed in.

Tools

The hosted server listed 68 tools to a connected client on Oct 16, 2026. The save_* tools both create and update: called without an id they create, and with one they edit.

Issues and triage

Labels

Projects, milestones and status updates

Documents and comments

Attachments and images

Releases

Diffs and code review

Linear's diffs are pull requests shown in Linear. These tools accept a Linear review URL or a GitHub PR URL.

Workspace, people and notifications

Set it up

Directly

Claude Code, using the command from Linear's docs:

claude mcp add --transport http linear-server https://mcp.linear.app/mcp

Run /mcp in a session to sign in. In Cursor, add the URL to ~/.cursor/mcp.json, or use the install link on Linear's page:

{
  "mcpServers": {
    "linear": {
      "url": "https://mcp.linear.app/mcp"
    }
  }
}

Linear offers a read-only endpoint, https://mcp.linear.app/mcp/readonly, that exposes only the read tools. You can also skip OAuth and send a Linear API key or OAuth token as Authorization: Bearer <token>. A key created with only the Read permission gives you a read-only connection. Each Linear workspace needs its own sign-in. The general client steps are in How to add MCP servers to Claude Code and How to set up MCP servers in Cursor.

Through Statio

Install Linear from the catalog, press Connect and approve on Linear's page. One person with permission to manage credentials connects it for the organization, and Statio refreshes the tokens before they expire. Then grant tools under Permissions to a person, a team or an agent. A team grant gives every member the whole server, so grant single tools to a person or agent when you want only some of them.

The desktop app writes the client config. By hand, point Cursor at http://127.0.0.1:7777/linear/mcp, using the server name from your dashboard, and for Claude Code run mcphub config claude-code. A non-interactive agent signs in with its own client ID and secret, which Connect a client describes. If nothing appears, start a fresh session and then work through How to debug an MCP server that won't connect.

What to hold back

A Linear connection is easy to think of as "the issue tracker". It also carries code review and merge rights over your GitHub pull requests, release management, and every team the user can see. Decide per agent.

A triage bot needs list_issues, get_issue, save_issue, save_comment, the label and status reads, and maybe get_triage_responsibility. Keep these from it:

• share_issue gives a person access to an issue they could not otherwise see. An agent following text in a bug report ("loop in alex@...") can expose an issue from a private team.
• save_status_update posts a project update that stakeholders read as the team's word. delete_status_update and delete_comment remove history.
• retire_issue_label on a label group retires every child label, which breaks any triage filter built on those labels.
• save_project, save_milestone and save_release move dates and stages on plans other people rely on.

A coding agent that picks up issues and opens pull requests should read diffs, and maybe post review comments. Hold back:

• merge_diff. It merges the pull request in your repository, or queues the merge. An agent that wrote the change and then merges it has skipped review entirely.
• submit_diff_review. An agent approving its own pull request is the same problem with an extra step.
• update_diff with its close and reopen actions, if you'd rather a person decides when a PR is abandoned.

get_notifications and mark_notification act on the connecting user's personal inbox. A shared agent identity marking things read hides them from the person who owns that inbox. Issue descriptions and comments are also untrusted input: anyone who can file an issue can write instructions an agent will read, which is why the write grants above matter. MCP permissions explains how per-tool grants work in Statio.

Questions