Linear earned its niche by treating engineers like adults. Keyboard shortcuts, fast UI, opinionated defaults. The integration with Claude Code earns its keep because Linear's data model is clean — tickets are predictable shapes, statuses are limited, projects are real.
The agent doesn't have to fight Linear the way it fights other tools.
Three flows that pay rent
Triage backlog. "Read all bugs labeled triage. Group them by root area, suggest a priority, flag duplicates." A bored engineering manager Sunday-evening task becomes a 5-minute review.
Cycle prep. "Pull all issues for next cycle. Flag ones with no estimate, no acceptance criteria, or unresolved comments." Walks you into Monday planning instead of out of it.
Issue from PR. "I just merged PR #1234. It fixes a bug we never had a Linear issue for. Create the issue retroactively for the audit trail."
Setup
Linear has a clean GraphQL API and a few MCP server options. We use the official linear-mcp.
npm install -g @linear/mcp-server
export LINEAR_API_KEY=lin_api_xxx
In claude_config.json:
{
"mcpServers": {
"linear": {
"command": "linear-mcp",
"env": {
"LINEAR_API_KEY": "..."
}
}
}
}
Verify with a small query: "List my Linear issues assigned to me, open." The agent should return them within a second.
Prompts that earn their keep
Cycle status, end of week. "Read all issues in the current cycle. Group by status. For each in-progress issue, summarize the latest comments. Output as a markdown table I can paste into our weekly update."
Standup draft. "Look at my Linear activity yesterday and my git commits. Draft my standup post: what I did, what I'm doing today, blockers. Don't make anything up."
Duplicate sweep. "Find Linear issues opened in the last 14 days with similar titles or descriptions. Group them. Suggest which is the canonical issue."
Refinement check. "List issues in the backlog state that have no estimate, no labels, or no description. Output as a checklist."
What to keep manual
- State transitions to
done. The agent can draft a closing comment. The button is yours. - Priority changes. Priority is a social signal across the team. The agent can suggest; you push.
- Reassignments. People's workload is a human conversation.
Caching and rate limits
Linear's API is fast but rate-limited. The MCP server caches reads for short windows. Don't fight it — the cache freshness is fine for most prompts.
For high-frequency flows (every-commit hooks, every-PR hooks), run a small local proxy that batches and dedupes Linear calls. Two lines of plumbing, big rate-limit relief.
What audit looks like
Every Linear comment the agent posts: tag via:claude-code. Filter on it weekly. If 20% of comments are agent-authored and they're all good, you have a healthy integration. If they're noisy, tune the prompt and raise the threshold.
Close
The integrations that compound are the ones that don't ask you to change your workflow — they just remove typing. Linear + Claude Code is that. Setup is 15 minutes. Time saved is in hours per week. The agent earns rent fast.
Related reading
- Claude Code + Jira — the parallel for non-Linear teams.
- The AI productivity playbook — broader workflow this fits in.
- Effective MCP patterns — the patterns this relies on.
We help teams wire AI into the tools they already use. Get in touch to skip the trial-and-error.