How to Install the Linear MCP Skill in Claude Code (2026 Step-by-Step)

Claude Code will prompt you for the API key. Paste the key you got in Step 1 and hit Enter.

If the command above doesn't recognize "linear" as a known shortcut (older versions of Claude Code don't), add it manually with the URL form:

When it prompts for authentication, paste your token.

Step 4: Verify the connection

In Claude Code, run:

You should see "linear" in the list with a green status indicator. If it's red, the most common cause is a typo in the API key. Re-add and try again.

A quick sanity check: in Claude Code, type:

> List the 5 most recent issues in my Linear workspace

Claude should respond with a real list pulled live from Linear, not a hallucinated example. If it returns a numbered list of fake-sounding issues like "PROJ-1234: Update homepage", you're not connected yet. Re-check Step 3.

Step 5: (Optional) Install the linear-claude-skill on top

The MCP gives you raw access to Linear's API. The skill layer adds opinionated workflows on top: weekly status reports, sprint planning helpers, dependency analysis. It's worth installing once the MCP itself is stable.

Or browse the [linear-claude-skill repo](https://github.com/) first. Always read SKILL.md before installing third-party skills so you understand what permissions you're granting.

Step 6: Try the five most useful queries

Once you're connected, here are the queries that pay back the setup time within a week:

1. Daily standup prep: "Show me everything assigned to me in 'In Progress' status with a description of where each one is stuck" 2. Weekly status: "List all issues my team closed last week, grouped by project, with a 1-sentence summary each" 3. Sprint planning: "Pull all unassigned issues with priority High or Urgent, sorted by oldest first" 4. Dependency check: "Find any issues with 'blocked by' relationships where the blocker is still open" 5. Quick triage: "Show me all issues created in the last 24 hours with no assignee"

Step 7: Create issues from chat

Writing new issues is where the time savings show up. Instead of switching to Linear, finding the project, picking the right team, copying context, you can say:

> Create a Linear issue in the Web project: title "Mobile nav menu doesn't close on outside tap", priority Medium, assign to Sarah, label "bug". Add this transcript of the bug report as the description: [paste].

Claude will draft the issue, confirm the details with you, then create it. The whole flow takes 15 seconds versus 90 seconds in the UI.

Step 8: Update statuses in bulk

When you finish a chunk of work, you can move 6 issues at once:

> Move issues PROJ-201, PROJ-204, PROJ-209 to 'In Review' and assign them to the design team. For PROJ-211, PROJ-214, PROJ-217, move to 'Done' and add comment "Shipped in v2.4".

This is the workflow where the MCP starts paying for itself. Bulk updates in the Linear UI are slow; in Claude Code they're conversational.

Step 9: Set up project-specific guardrails

If you work across multiple projects, add a CLAUDE.md note in your repo so Claude knows the default project context. Example:

• Default Linear team: Web
• Default project: q2-rebuild
• Always set label "frontend" on issues created from this repo

Now Claude defaults to the right team and labels when you create issues from this codebase. Small touch, big consistency win.

Step 10: The three mistakes most people make

Mistake 1: Storing the API key in plaintext somewhere risky. Use Claude Code's built-in secure storage (the MCP config encrypts it locally). Don't paste it into a Slack DM "for reference."

Mistake 2: Granting MCP write access to a workspace you don't fully control. If you're new to Linear, ask your admin first. A wrong bulk update from an MCP is harder to undo than a wrong click in the UI.

Mistake 3: Skipping the SKILL.md read on community skills. Some community Linear skills hardcode behaviors you might not want (auto-closing stale issues, auto-assigning to specific people). Read the file before installing.

Step 11: Combine it with other MCPs

The Linear MCP gets dramatically more useful when chained with others:

• Linear + GitHub MCP: "When a Linear issue is moved to Done, find the matching PR in GitHub and confirm it's merged. Flag any mismatches."
• Linear + Notion MCP: "Pull every issue in the 'Q2 Launch' project and create a Notion page summarizing scope, blockers, and risks."
• Linear + Slack MCP: "Post a daily summary of issues moved to 'In Review' to #design-reviews."

This is where the real workflow gains show up. Single-MCP usage is helpful; multi-MCP orchestration is the unlock.

Step 12: When to remove it

If you stop using Linear, remove the MCP cleanly:

This deletes the local config and revokes Claude Code's access. Then go to Linear settings and revoke the API key on Linear's side too. Belt and suspenders.

My honest take after using this for 3 months

The Linear MCP is one of the highest-ROI MCPs you can install in 2026 if you're already a heavy Linear user. The first week feels novel. The second week it becomes "how did I work without this." The bulk update workflow alone justifies the setup time.

It's also worth saying: this isn't a replacement for the Linear UI for everyone on your team. Your designers and product managers will still want the visual sprint board. The MCP is for the people whose work surface is already a terminal or a Claude conversation.

FAQ

Is it free? The MCP itself is free. You're using your existing Linear plan and Claude Code's existing model usage. No additional subscription.

Does it work with Linear's free plan? Yes. Any Linear plan with API access works (free plan included as of 2026).

Can I limit what the MCP can do? Yes. Linear API keys can be scoped. For an MCP that only reads (no writes), create a read-only API key in Linear settings. Safer for teams that just want reporting.

Is my Linear data sent to Claude's servers? Yes, when Claude needs to reason over it. If you have strict data residency requirements, use Claude Enterprise with a configured data retention policy, or run a self-hosted MCP server with your own model.