Linear MCP: What It Actually Does (and How to Set It Up)

Linear MCP is Linear's official Model Context Protocol server, a remote server Linear hosts so AI clients like Claude, Cursor, ChatGPT and VS Code can find, create and update Linear issues, projects, comments and other objects while signed in as you. Linear launched it on May 1, 2025, its docs do not label it beta, and its pricing page lists MCP access on every plan, including Free.

I read Linear's docs, changelog and pricing, plus the setup guides from Anthropic, OpenAI and Cursor, on October 6, 2026. Where vendor pages disagree, I say so.

What is the Linear MCP server?

Linear's MCP docs describe a centrally hosted server that follows the authenticated remote MCP spec over Streamable HTTP.

Older guides, including Linear's 2025 launch post, use `https://mcp.linear.app/sse`. Linear announced on February 5, 2026 that it was removing SSE, and that address returned a 404 when I tried it on October 6, 2026, though the docs FAQ still calls it a deprecated fallback. Use `/mcp`.

Separately, Linear Agent can call other MCP servers, such as Glean or Notion, since April 23, 2026. That is the reverse direction; this post covers Linear data going to your AI tools.

What can the Linear MCP server do?

Linear's docs describe tools for finding, creating and updating issues, projects, comments and more, but publish no tool list. The fullest public list I found is Anthropic's directory entry for the connector, which Linear built. It names 22 tools, and Linear's changelog records changes since:

• February 5, 2026: tools to create and edit initiatives, project milestones and project or initiative updates, and to manage project labels.
• February 26, 2026: create_issue and update_issue became one save_issue tool. The May 14 changelog refers to save_project and save_document.
• July 2, 2026: tools for releases and release notes.

Check your client's tool list after connecting.

Writes are on by default. To go read-only, connect to `/mcp/readonly`, request only the `read` OAuth scope on `/mcp`, or use a Linear API key with only the Read permission. The server's OAuth metadata confirms the read-only resource offers only the `read` scope.

How do you set up Linear MCP in Claude, ChatGPT and Cursor?

Claude (claude.ai and desktop)

Add the Linear connector under Connectors in Claude's settings. Anthropic's directory lists it as Anthropic verified and made by Linear.

On Team and Enterprise plans, an Owner first adds it under Organization settings, Connectors, then each person signs in to Linear (Anthropic). Owners can mark each tool Always allow, Needs approval or Blocked; Anthropic's own example lets Claude view Linear issues but not create them or change status.

Linear Enterprise workspaces on Okta can skip per-person sign-in. With Okta SAML set up, an admin turns on enterprise-managed authorization in Linear, and Linear then verifies employees through Okta when they connect from Claude, applying their existing permissions. Anthropic lists enterprise-managed auth as beta.

Claude Code

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

Then run `/mcp` in a session to sign in. Servers default to local scope (the current project only), so add `--scope user` to use Linear everywhere (Claude Code docs).

ChatGPT

Linear documents a ChatGPT connector for deep research, connected from ChatGPT's settings and listed for Plus and Pro users outside the EEA, Switzerland and the UK, and for Team, Enterprise and Edu. OpenAI's help center now manages these as connected apps under Settings, Plugins, so menus may differ from Linear's page.

For the MCP server itself, with writes, use a custom MCP app. OpenAI's developer mode guide lists this for ChatGPT Business, Enterprise and Edu on the web, with full MCP support in beta. An admin enables developer mode, creates an app with Linear's endpoint, picks OAuth, runs Scan Tools, tests it and publishes it. Linear's docs do not cover this route, so pilot it first.

Cursor

Use the one-click link in Linear's docs, or install the Linear plugin from Cursor's marketplace. It is created by Linear, verified by Cursor, and its config points at `https://mcp.linear.app/mcp`. Marketplace installs sign you in with OAuth (Cursor docs). To add it by hand, put this in `~/.cursor/mcp.json`, or `.cursor/mcp.json` for one project:

{   "mcpServers": {     "linear": {       "url": "https://mcp.linear.app/mcp"     }   } }

Cursor asks before running MCP tools by default.

Other clients

For VS Code, Windsurf and Zed, Linear's docs give an mcp-remote config, which also covers clients without remote MCP support: `npx -y mcp-remote https://mcp.linear.app/mcp`.

How do Linear MCP's authentication, permissions and limits work?

OAuth in the browser is the default. Linear also accepts an API key or OAuth token in an `Authorization: Bearer` header, for example to act as an app user or to give read-only access through a restricted key. Admins decide whether Members can create API keys, and each key can be limited by permission (such as Read only) and by team (Linear API docs).

The server acts as whoever signed in; Anthropic notes that Claude inherits each person's permissions from the connected service.

Linear's MCP docs list no rate limits. Its GraphQL API allows 2,500 requests per hour per user with an API key and 5,000 with an OAuth app; the docs do not say whether MCP calls count against those. Linear's integration pages also warn that remote connections may fail or need several attempts, and an April 2026 changelog entry fixed OAuth connections that could drop after about a day.

Where does a single-vendor MCP server stop?

Linear's server handles Linear well. The limits show up when a question is bigger than Linear.

It reaches one app

"Which bugs customers raised in Zendesk this month have no Linear issue, and what did engineering say in Slack?" needs Zendesk, Slack and Linear servers at once. The model joins the results itself, from scratch, every time someone asks.

Every server adds to the context window

Tool definitions and results both use context. Anthropic suggests On demand tool loading once 10 or more connectors are active, and Claude Code warns when one MCP result passes 10,000 tokens. Clients are improving, but cross-tool questions still pull results from several servers into one context.

Each person and team wires its own

Enabling a connector for a Claude organization signs nobody in, and each Linear workspace needs its own sign-in. In ChatGPT Business, only admins can build and publish the app, and changing it means recreating it. Okta-managed authorization removes the per-person step only for Linear Enterprise on Okta with Claude.

If you want outside context inside Linear itself, test Linear Agent's MCP connections first.

Linear MCP vs Coworker MCP: how do they compare?

They solve different problems, so the useful comparison is scope.

*Statistically significant benchmarks comparing Coworker MCP vs. Claude Native Tooling: 100 tasks across 7 categories. The 9x and 64% figures are the best case (retrieval-heavy categories, with Coworker Learning); across all 7 categories the same setup was 75.5% cheaper and 45.6% faster, and blind reviewers preferred its answers 84.5% of the time. Linear was not among the nine sources tested (benchmarks).

What should you check before connecting Linear MCP?

1. Decide who gets writes. Put most people on `/mcp/readonly` or the `read` scope, and give write access to those who triage and plan.
2. Require approval for writes. Mark write tools Needs approval in Claude Team and Enterprise; Cursor asks by default. Several of Linear's example prompts ask the model to preview changes before applying them.
3. Prefer per-person OAuth to shared API keys, and limit any key to Read or to specific teams.
4. Check where data goes next. OpenAI says it may train on information from apps for Free, Plus, Go and Pro users with "Improve the model for everyone" on, but not by default for Business, Enterprise and Edu.
5. Treat issue text as untrusted. Issues from customer intake carry outside text, so keep write approvals on (MCP security).
6. Retire old configs. Replace `/sse` URLs and community servers; the early jerhadf/linear-mcp-server README now says it is deprecated and points to Linear's server.

For a wider rollout plan, see our guide to enterprise MCP.

Where does Coworker MCP fit?

If you only need Linear inside Claude or Cursor, use Linear's server. Linear builds it and includes it on every plan, and its changelog shows MCP changes in most months since launch.

Coworker MCP is for questions that cross tools: one connection to 50+ tools, including Linear, Slack, GitHub, Zendesk, Intercom and Gong, backed for large organizations by OM2, Coworker's organizational memory. "You can think of OM2 as a context graph," Coworker's CEO wrote in the OM2 launch post. Documents, messages and tickets become atomic facts linked to entities and relationships, ranked by proximity, recency, entity centrality and relationship strength. Per the launch post, structured queries over that graph handle counting, negation and multi-hop traversal, with source permissions enforced on traversal.

Per the Coworker MCP page, an admin enables MCP access, people sign in with their existing Coworker login (SSO supported), and setup typically takes under 30 minutes. The trade-off is breadth: 50+ connectors, not every tool you use. For a similar comparison, see Glean MCP vs Coworker MCP.

If your questions start in Linear and end in Slack, support tickets or call notes, book a demo to see Coworker MCP answer them in Claude or Cursor, or request a Context Impact Report on your own data.

Frequently asked questions

Does Linear have an official MCP server?

Yes. Linear launched its official remote MCP server on May 1, 2025. It lives at https://mcp.linear.app/mcp and connects Claude, Cursor, ChatGPT and other MCP clients to your Linear workspace.

What is the Linear MCP server URL?

The main endpoint is https://mcp.linear.app/mcp, read-write by default, and https://mcp.linear.app/mcp/readonly is read-only. Linear announced the removal of the old /sse endpoint in February 2026, and it returned a 404 on October 6, 2026.

Is the Linear MCP server free?

Linear's pricing page lists MCP access on every plan, including Free. Your AI client may add requirements: OpenAI's guide lists ChatGPT's custom MCP apps for Business, Enterprise and Edu workspaces.

How do I add Linear MCP to Cursor?

Use the one-click link in Linear's docs or the Linear plugin in Cursor's marketplace, which signs you in with OAuth. Manually, add a server named linear with the URL https://mcp.linear.app/mcp to ~/.cursor/mcp.json.

Can Linear MCP be read-only?

Yes. Connect to https://mcp.linear.app/mcp/readonly, request only the read OAuth scope, or use a Linear API key with only the Read permission. Claude Team and Enterprise owners can also block or require approval for write tools.

Does Linear MCP work with ChatGPT?

Yes. Linear documents a ChatGPT connector for deep research. For the MCP server itself, with write actions, ChatGPT Business, Enterprise and Edu admins can add it as a custom app in developer mode, which OpenAI describes as beta and web-only.

Is there an open-source Linear MCP server?

Community servers exist on GitHub, but Linear's official server is hosted, and I found no source for it among Linear's 22 public GitHub repositories. The early community server jerhadf/linear-mcp-server now says it is deprecated.

What is the difference between Linear MCP and Coworker MCP?

Linear MCP reaches only Linear and is the simpler choice for Linear work in one AI client. Coworker MCP is one connection to 50+ tools, Linear included, backed for large organizations by OM2 organizational memory, for questions that span tools.

Related reading

• Coworker MCP
• Linear connector
• What is MCP?
• MCP security
• Claude MCP
• ChatGPT MCP
• Enterprise MCP
• HubSpot MCP
• Glean MCP vs Coworker MCP
• What is a context layer?
• Organizational memory (OM2)