Linear + Claude Code: How to Connect Them Today (and the Shortcut)

Linear has leaned hard into agents: its agent directory lets you assign issues directly to Cursor, OpenAI Codex, Devin, and others, and for those tools the loop is genuinely native, delegate an issue in Linear, get work back. Claude Code is not in that directory as of July 2026. There is an open feature request for the integration on the Claude Code GitHub repo, which tells you two things: demand exists, and nothing official has shipped.

So teams running Claude Code, by many accounts one of the most widely used coding agents, face an odd asymmetry: their tracker speaks natively to their agent's competitors, but not to their agent. Everything that follows is about closing that gap with what exists today. The general pattern, issue to agent to PR, across trackers is covered in how to assign tasks to AI agents.

Linear ships an official MCP server, and Claude Code speaks MCP natively. Connecting the two gives every Claude Code session read and write access to your Linear workspace: the session can pull an issue's description and comments as context, update its status, and comment with progress. It is the lowest-friction option because it uses only first-party pieces, no third-party service in the middle.

1. 1

   Add Linear's MCP server to Claude Code

   Register the server in your Claude Code MCP configuration, the same way you add any MCP server. Linear's docs cover the current endpoint and setup; our MCP guide covers how Claude Code manages server configs.
2. 2

   Authenticate with your Linear account

   On first use you authorize the connection against your Linear workspace, so sessions act with your permissions, they see the teams and issues you see.
3. 3

   Reference issue IDs in your prompts

   From then on, prompts like "implement ENG-142, then mark it In Review and summarize what you changed in a comment" work: the session fetches the issue itself and writes its updates back.

The honest limit: this is context, not delegation. The session knows about Linear, but you still open every session, supervise it, and manage branches yourself, and nothing in Linear shows that an agent is working the issue beyond the comments it leaves. Setup details for MCP generally live in MCP servers for AI coding.

If you want real delegation, assign an issue in Linear, agent picks it up, the missing native integration can be approximated by a bridge. Cyrus is the notable one: open source, it listens for Linear issues delegated to it and runs Claude Code against them, posting progress back to the issue. Functionally, it is the closest thing to Claude Code being in Linear's agent directory today.

The third option questions the premise. Options A and B work to connect a tracker that doesn't know about Claude Code to an agent that doesn't know about Linear. AIDEN skips the bridge entirely by putting the board and the agents in one app: stories live on a kanban board, each gets a written spec you approve, and your Claude Code (or Codex) CLI executes each one on its own worktree, with the diff reviewed on the same card. Nothing to glue, because there is no seam.

In practice this is rarely Linear-versus-AIDEN. Teams keep Linear for what it is genuinely best at, company-level planning: roadmaps, cycles, cross-functional visibility, while the agent lane runs on AIDEN's board: a high-priority Linear item becomes an AIDEN story, an agent implements it behind the spec gate, the PR merges, the Linear issue closes. The tracker plans; the board executes. When a team does want one system, the tradeoffs are laid out in Linear alternative for AI teams, and why agent work wants a tracker shaped differently from a human one is the subject of issue tracking for AI agents.

What the execution side actually needs, specs as first-class objects, a worktree per story, sessions as visible cards, review as a column, is exactly what issue comments can't carry, and it is covered in Claude Code orchestration. AIDEN is free for one project, so running the Linear-plans, AIDEN-executes split costs nothing to try.