← mcp directory● official

MCP server · Linear

Linear.

When the issue tracker becomes a tool the model can call, agentic workflows close the loop end-to-end. Read the bug, fix it, comment with the diff, mark it done.

$ claude mcp add linear -- npx -y @linear/mcp

Most agentic workflows die at the seam between code and project management. The agent fixes a bug; the developer still has to copy the fix into a Linear comment, change the status, and ping the reporter. The Linear MCP server makes the whole loop machine-readable, so the seam disappears.

You hand the agent a Linear issue ID. It reads the issue, checks the linked PR, makes the fix, comments with a summary, and updates the status, all in one continuous turn. The unglamorous truth is that this is what most so-called "AI agents" should be doing: closing loops between systems that already exist, rather than inventing new systems that need their own loops.

§01What you get

The server exposes the operations you would expect from the Linear API, scoped as MCP tools:

  • Read issues, comments, projects, cycles, and users.
  • Create and update issues, including title, description, status, priority, assignee, and labels.
  • Comment on issues programmatically.
  • Search using Linear's filter syntax (assignee, label, state, and so on).
  • Transition issues across workflow states from a prompt.

§02When it earns its keep

The pattern that pays off is a standing prompt that runs on a schedule, scoped to a small set of issues. The kind of prompt that used to be a 15-minute ritual:

◆ pull quote

“Find every P0 bug filed today that is unassigned. For each, read the description, decide if it is frontend or backend, and assign it to the relevant on-call.”

As an MCP-driven prompt scheduled via cron, that is a 30-second job that runs while you are doing something else. The value of MCP servers like this one is that they collapse the cost of a recurring decision down to a single prompt invocation.

§03Setup

# you'll need a Linear API key with the
# right scopes. create one at:
# linear.app/settings/api

export LINEAR_API_KEY=lin_api_xxxxxxxx

claude mcp add linear -- npx -y @linear/mcp

claude mcp list

If you are using a manual JSON config, drop the env var into the server entry under env rather than leaving it floating in your shell.

§04Caveats: this one writes

  • Write access is real. The agent can change status, reassign, comment publicly, and (depending on scope) delete. Default to a read-only key for exploratory work.
  • Scope per environment. A key that can read your team's Linear is a leak risk. Prefer per-workflow keys with the narrowest scope you can stomach.
  • Comments are visible to the team. An agent that comments on an issue is an agent your colleagues will see. Tone matters; the system prompt should make clear what voice the agent uses.
  • Rate limits exist. Linear's API is generous but not infinite. A misconfigured loop will get throttled before it gets expensive.