Linear Claude Connector: The Complete Guide

Once set up, Claude agents can create issues, update statuses, manage projects and search across your Linear workspace without leaving your client. Linear's MCP server follows the authenticated remote MCP spec, so the server is centrally hosted and managed - you don't run it yourself.

• Create and update issues directly from your IDE or chat.
• Move issues through status workflows and assign to teammates.
• Search issues by project, label, assignee or other filters.
• Manage projects - create, update, archive and unarchive.
• Read sprint data via list_cycles for summaries and risk flags.
• Resolve comments - track whether feedback was addressed.

Linear has long offered a powerful GraphQL API and webhooks, and that remains the right tool when you are building a custom integration, a dashboard, or an automated pipeline that runs without a human in the loop. You manage a personal API key or an OAuth application, write queries against the schema, and own the plumbing.

The MCP server is a different proposition. It is not a replacement for the GraphQL API - it is a managed, conversational front door designed for AI agents. Linear hosts it centrally, exposes a curated set of tools (create, update, search, list cycles, comment), and handles auth through an interactive flow, so Claude can act on your workspace without you writing a single GraphQL query. For the "log this bug and assign it to me" use case, MCP is dramatically less work; for a nightly sync into your data warehouse, the API still wins. Most teams end up using both: the API for systems integration, MCP for the in-editor agent loop.

• In-flow ticket creation: while working on code, ask Claude to "log this as a bug, assign to me, label 'auth'". Done without context switch.
• Issue triage: an agent analyses descriptions, checks for similar existing issues, sets priority and routes to the right team - reducing the time engineering managers spend triaging.
• Sprint summary: "Summarise the current cycle - what's on track, at risk and overdue?"
• Bulk updates: reassign all issues with a given label, or close stale tickets in one pass.
• Feedback follow-through: after a review, check that all the comments have been resolved.

The loop Linear users cite most often - logging and tracking a bug without ever leaving the editor:

1. While debugging in Claude Code you spot a real issue. You say: "Log this as a bug in the Web team, title it 'Session token not refreshing', assign it to me, label it 'auth' and set priority high."
2. Claude creates the issue with all of that metadata in one call and returns the issue identifier.
3. You keep coding. When you push the fix, you ask: "Move WEB-412 to In Review and link this PR."
4. Claude updates the status and adds the link.
5. At standup you ask: "Summarise the current cycle for the Web team - what's on track, at risk and overdue?" - Claude reads list_cycles and produces the summary.
6. After review you ask Claude to confirm all the review comments on the issue were resolved before closing it.

No tab-switch to the Linear board, no manual form-filling, no losing the thread of the code you were writing. The agent keeps the tracker honest while you stay in flow - which is the entire design philosophy of Linear extended to the AI layer.

Linear's user base skews towards startups and fast-moving teams who value flow, and the connector reflects that. Write-ups by Merge.dev, Composio, Verdent, Murphy Randle ("Freeing Up Flow With Claude Code & Linear MCP") and others repeat the same theme: ticket admin stops breaking flow. Pair the connector with Claude Opus 4.8 and Claude Code, and the loop from "found a bug" to "logged and assigned" collapses to seconds.

Linear runs a hosted remote MCP server at https://mcp.linear.app/mcp. There is nothing to install or self-host - you register the URL with your client and complete an OAuth 2.1 flow with dynamic client registration.

Claude Code

Add the server, then trigger the auth flow:

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

Open a Claude Code session and run /mcp to start the OAuth flow. A browser window (or Magic Link) opens; approve access and you are connected. Run claude mcp list to confirm the server is registered.

Claude Desktop and claude.ai

On Claude Team and Enterprise, go to Settings → Integrations (or Connectors), click Add more, and enter the name Linear with the URL https://mcp.linear.app/mcp. Complete the OAuth consent and the Linear tools appear.

Other clients via mcp-remote

For clients that don't yet speak remote MCP natively, bridge with the mcp-remote proxy in your config file:

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.linear.app/mcp"]
    }
  }
}

Restart the client, complete OAuth in the browser window it opens, and start issuing natural-language commands to manage issues, projects and cycles.

Authentication happens via the Magic Link, and the connector operates with your existing Linear permissions - it can only see and act on workspaces and projects you already have access to. The standard advice applies: scope tightly during pilot, watch what writes the agent makes, and prefer narrow tasks while you build trust.

The Linear connector is one of the more reliable in the directory, but a handful of issues recur in setup threads.

Tools stop responding mid-session

The most common complaint, and the one with the simplest fix. Linear's own guidance: restart your client or disable and re-enable the Linear MCP server. The remote server occasionally drops a session token; a clean reconnect re-runs OAuth and restores the tools. If a single command hangs, cancel and reissue rather than waiting.

OAuth window never completes

The callback runs on localhost. VPNs, strict corporate proxies or browser extensions that block localhost redirects will stall the handshake. Complete the consent in a standard browser session, and in Claude Desktop make sure npx is on PATH so mcp-remote can launch.

Agent can't find an issue or team

The connector only sees what your Linear account sees. If a search returns nothing, confirm you are a member of the relevant team - Linear scopes issues by team, so a private team you have not joined is invisible to the agent. Name the team explicitly in the prompt to disambiguate when you belong to many.

Bulk writes go too far

"Close all stale tickets" is fast to say and fast to regret. Ask Claude to list the issues it intends to change before it runs the update, and scope by label and team. Linear keeps a history, but undoing a bad bulk reassignment by hand is tedious.

Wrong workspace after switching accounts

If you belong to multiple Linear workspaces, the connector authenticates against the one you consented to. To switch, disable and re-enable the server and pick the right workspace during the OAuth step.

The MCP server is included with Linear at no additional cost - any plan that gives you workspace access gives you the connector. There is no separate per-seat charge for connecting Claude, and the actions the agent takes count against your existing workspace exactly as if you performed them yourself.

The server is generally available and centrally hosted by Linear, so there is no version to keep up to date and no infrastructure to run. Tool availability tracks your Linear plan's features - for example, cycles (sprints) must be enabled on the workspace for list_cycles to return data.

• Workspace-bounded: the connector sees what your account sees, no more.
• Write impact: bulk updates can be quick to do and quick to regret - review before running broadly.
• Linear-shaped: excellent inside Linear, not a cross-tool orchestrator.

Engineering teams on Linear - especially those using Claude Code. The whole point of Linear is keeping engineers in flow; the connector extends that philosophy by letting Claude handle the ticketing without breaking it.

What does it do?

Create and update issues, manage projects, move issues through workflows, read sprint data, resolve comments.

Is it official?

Yes - Linear documents the MCP server in its docs and integrates with Claude natively.

How do I authenticate?

A Magic Link appears the first time you invoke a Linear tool.

What if it breaks?

Restart your client or disable and re-enable the Linear MCP server.

The Linear connector lets engineers keep shipping while the agent keeps the tracker honest. Issues created, status updated, sprint summarised - all without leaving the editor. For Linear-first teams, especially those using Claude Code, it is one of the cleanest productivity wins in the directory.

Connect with the Magic Link, start with single-issue tasks, and expand as you go. Explore more in the complete Claude Connectors Directory.