Set up Linear MCP Server

The Linear MCP Server connects your AI agent directly to your Linear workspace: find, create, and update issues, get project overviews, and review cycles — all from within your AI client. Since the server has full write access, a well-thought-out permissions setup is recommended from the start.

Establish the connection

For Claude Code, a single command is enough: claude mcp add --transport http linear-server https://mcp.linear.app/mcp. Then enter /mcp in a Claude Code session to start the OAuth authentication flow. Other clients (Cursor, VS Code, Windsurf, Zed, Codex) are connected via their MCP configuration file using npx -y mcp-remote https://mcp.linear.app/mcp. On first connection, a browser window opens for the Linear OAuth login.

Choose read vs. write access

For analyses, status reports, or standup summaries, the readonly endpoint https://mcp.linear.app/mcp/readonly is recommended. It exposes only read tools according to the provider, structurally ruling out accidental write access. The default endpoint https://mcp.linear.app/mcp offers full read and write access — useful when the agent should also create or update issues.

Use tools deliberately

Start with clear, tightly scoped prompts: "Show me all open issues assigned to me in Cycle 12" or "Create an issue in Project X with title Y and assign it to me". For more complex automations — such as automatically creating issues from meeting notes — a test run in a development workspace is recommended before deploying the agent in the production workspace.

Keep security in mind

Since the server exposes write tools, it should only be set up in trusted client environments. Prompt injection via Linear content (e.g. issue descriptions, comments) is a general risk; the agent should not execute instructions from these fields directly without validating them. Enterprise teams can additionally secure access via Okta SAML.