Linear

Limits

• The connector posts comments that notify real colleagues immediately. Linear sends desktop, mobile and Slack notifications in real time to every subscriber, and email on the recipient's own schedule. Linear documents no confirmation step; your MCP client is the only place an approval gate can exist.
• The published tool counts disagree: 22 against 53. Every delete tool and the merge_diff tool appear only on the larger list, and the directory's create and update tools have been renamed to save_* on the live server. A reassurance derived from the 22 names is unsafe.
• Rate limits are real budgets and MCP calls spend them. Linear meters its GraphQL API on two independent axes per user per hour. An OAuth app gets 5,000 requests and 2,000,000 complexity points; an API key gets 2,500 requests and 3,000,000 points; a single query is hard-capped at 10,000 complexity points and rejected above it.
• Complexity is charged on data volume, not on call count. Linear prices each property at 0.1 points, each object at 1, and multiplies a connection's children by the pagination argument — defaulting to 50 when unspecified. Its own worked example puts a plain "list my created issues with three fields" query at 66 points. An agent that lists broadly without pagination arguments spends the budget far faster than the request counter suggests.
• Linear's own rate-limit page contradicts itself on the API-key request limit. Its prose says 5,000 requests per hour for API keys; the table on the same page says 2,500 for API keys and 5,000 for OAuth apps. We report both and do not resolve it — settling it would require authenticating, which we never do.
• Rate-limit errors return HTTP 400, not 429. Linear states GraphQL rate-limit responses carry a 200-class HTTP shape with a RATELIMITED code in the errors body. A client that retries only on 429 will not back off correctly.
• Guests inherit workspace integrations. Linear warns that integrations enabled for a workspace are accessible to guest users, and that this could allow access to data from teams they were not invited to. It recommends reviewing third-party integrations individually.
• Audit logging is Enterprise-only and retained 90 days. Linear's audit log is available on the Enterprise plan, tracks account access and settings changes with actor IP and country, and retains entries for 90 days.
• Third-party app approval is Enterprise-only too. The workspace-level control that would let an owner gate this connector before members install it requires the Enterprise plan.
• The SSE endpoint is deprecated. Linear directs all new setups to Streamable HTTP and keeps /sse only as a fallback — including for WSL on Windows, where it is still the documented workaround.
• We did not read the tools and never called one. Our only contact was an anonymous handshake that returned HTTP 401 and four unauthenticated metadata fetches. Every statement about what a tool does comes from Linear's documentation or Anthropic's directory listing, each dated below.

Frequently asked questions