How to Connect Claude Code to Linear With the Official MCP Server

Connect Claude Code to Linear by adding Linear's own hosted MCP server, at https://mcp.linear.app/mcp, and authenticating once through your browser. Once it's connected, Claude Code can search your Linear workspace, pull an issue's full description and comment history as context, and create or update issues, change status, assign people, and leave comments, all from inside the same session where you're already building.

This is a server Linear built and runs themselves, not a community package reverse-engineered against their API. That matters the same way it did for [connecting Claude Code to Notion](/blog/connect-claude-code-to-notion): it's maintained against Linear's actual product, so it keeps working when Linear ships changes instead of quietly breaking. Community alternatives exist too, but the hosted route is the one I'd default to unless you have a specific reason not to.

The hosted route is the fastest, and it's the one Linear documents as the default.

If you'd rather run a narrower, read-only connection, Linear lets you generate an API key with only the Read permission enabled, then point a local MCP server at that key instead of using the hosted OAuth route. That trade gives up write access, so Claude Code can pull issue context but can't create or update anything on your behalf, which is the right call for a shared machine or a client-facing workspace you'd rather not risk.

The obvious use is starting a build from the ticket instead of retyping it. Point Claude Code at a Linear issue and it reads the real description and every comment, including the back-and-forth a client or teammate already had about scope, instead of you summarizing that thread into a prompt by hand and losing detail on the way.

I run this as a loop, and inside the club we just call it the Ticket-to-Diff loop: pull the issue, build against exactly what it says, then push the result back into the same issue instead of leaving it stranded in your terminal history. It sounds small, but it's the difference between Linear being a to-do list you check before you code and Linear being the actual system of record for what happened.

Linear's MCP connection doesn't have its own permission model separate from you. It authenticates as your Linear account and inherits exactly what you can already see and edit, team by team. If you can't access a team in Linear's own interface, Claude Code can't reach it through the connection either, and if you can, it can, so the real access-control decision already happened when your workspace admin set up your account.

That makes the scope conversation different from Notion's page-by-page sharing model: there's no separate list of what's shared with the integration to review. Workspace admins can still restrict MCP access at the org level with allowlists if that's a concern, but for most solo builders and small agency teams, the thing to actually check is whether your own Linear seat has access to more teams than the client work in front of you needs.

Skip it for a one-off build with no Linear presence, or a client who tracks work in a spreadsheet or a shared doc instead. It earns its place once there's a real, actively-updated issue tracker on the other end, where skipping the connection means you'd be retyping ticket context into a prompt every session anyway.