Linear MCP Server: Tools, Setup, and the SSE Retirement
Yes — Linear runs an official MCP server, and it’s the one you want. It’s remote and hosted by Linear at https://mcp.linear.app/mcp, there’s nothing to install, and authentication is OAuth 2.1 with dynamic client registration — the first connection opens a browser, you approve, and the client reuses the credentials. Linear shipped it on May 1, 2025, following what it called the “authenticated remote MCP spec,” with the server “centrally hosted and managed.”
The reason this matters more for Linear than for most apps: the clients people connect it to are coding agents. Claude Code, Codex, and Cursor sitting inside the repo while also holding the ticket is the actual use case, not a novelty. What follows is what the server documents itself as exposing, the setup line for each client, the transport change that’s currently breaking older configs — and the one class of work it structurally cannot do.
Linear is one of many — the complete list of MCP servers covers every app that ships one, grouped by category.
What the Linear MCP server exposes
Linear’s own MCP docs describe the scope in one sentence: the server “has tools available for finding, creating, and updating objects in Linear like issues, projects, and comments — with more functionality on the way.”
That’s deliberately object-shaped rather than a tool list, and it’s worth taking literally instead of trusting the inflated counts floating around third-party directories. Read against Linear’s own changelog, the documented surface is:
- Issues — find, filter, create, update. Status changes, assignment, and labels all sit here.
- Projects — find, create, update.
- Comments — read and post, which is what makes an agent’s work visible to the rest of the team rather than trapped in your terminal.
- Initiatives and initiative updates — added February 5, 2026, when Linear extended the server past pure issue tracking.
- Project milestones, project updates, and project labels — same February release.
- Images and URL-based resource loading — the server can load images, and you can point a client at a Linear URL to pull that resource in directly. Pasting an issue link into a chat and having the agent fetch it is a real workflow, not a hack.
The February change was explicitly aimed at product managers rather than engineers — the pitch was maintaining plans and communicating progress from Cursor or Claude without opening Linear. It also came with what Linear described as better tool documentation, which in practice means the model picks the right tool more often and burns fewer turns guessing.
How to connect it, per client
There’s one endpoint and roughly one line per client.
Claude Code:
claude mcp add --transport http linear-server https://mcp.linear.app/mcpCodex:
codex mcp add linear --url https://mcp.linear.app/mcpCodex needs its experimental RMCP support enabled for remote servers. Our Codex MCP servers guide covers the config surface around that — auth modes, timeouts, and the enabled_tools allowlist that’s worth applying to a server this broad.
Claude (web and desktop): it’s a first-party connector. Add Linear from connector settings; no URL to paste. Enterprise workspaces on Okta SAML are supported for Claude clients.
Cursor: install via Linear’s direct link or find it in Cursor’s MCP directory.
VS Code, Zed, Windsurf: these still go through the mcp-remote wrapper:
npx -y mcp-remote https://mcp.linear.app/mcpChatGPT: connects as a remote MCP-backed app rather than through any local config — see ChatGPT MCP servers for how that surface differs from the CLI clients.
Beyond the interactive OAuth flow, the server accepts an OAuth token or API key passed straight through as Authorization: Bearer <yourtoken>, which is the path for app users, service accounts, and CI. Access itself isn’t a separate purchase — you’re authorizing a client against the Linear workspace and permissions you already have, and admins can constrain it with allowlists and workspace-level MCP permissions.
The SSE endpoint is being retired
If you’re troubleshooting a config that used to work, check this first. Linear’s original endpoint was https://mcp.linear.app/sse, and as of the February 2026 release it’s being phased out in favor of https://mcp.linear.app/mcp, which uses Streamable HTTP. Linear described the transition as gradual, over roughly two months. Anything still pointed at /sse should move.
Two other failure modes worth knowing: stale credentials usually clear with rm -rf ~/.mcp-auth, and WSL users generally need the --transport sse-only flag to get a connection at all.
”Linear MCP” also means the opposite thing now
This trips people up in search results. Since April 23, 2026, Linear Agent can consume external MCP servers — you connect Granola, Notion, or PostHog in agent personalization settings or triage automations, and the agent inside Linear reaches out to them. Linear’s example is pulling meeting takeaways from Granola into a project update and creating follow-up issues.
So there are two directions with the same name: your AI client reaching into Linear (the hosted server at mcp.linear.app/mcp), and Linear’s own agent reaching out to other apps. Most “linear mcp” searches want the first one. If you wanted the second, it’s configured entirely inside Linear and admins gate it with the same allowlists.
Where the Linear MCP stops
The server is well built and Linear maintains it properly. The limits are the protocol’s, not Linear’s:
- No triggers.
MCP servers wait to be asked. The protocol has no scheduler and no event mechanism — its maintainers acknowledge that change detection today means polling — so an MCP connection can answer a question but cannot notice something on its own.
An issue moving to In Review, a cycle ending, a P0 landing in triage — none of these can start anything. - One app per connection. The Linear server knows Linear. A bug that should create the issue, page the on-call, and post in Slack means a separate server, separate auth, and a client that can hold all of them in one turn.
- Every tool costs context. A broad server loads its full tool list into every turn. Trimming to the handful you actually use is the difference between a snappy agent and one that spends its budget on menu descriptions.
- The blast radius is yours. OAuth grants under your account permissions. Auth agents with read-only or service-account scopes rather than your admin session when they don’t need writes.
None of that is a problem when you’re at the keyboard — an agent reaching into Linear mid-task is exactly the point. It becomes a problem for everything that happens when you aren’t.
Making Linear react to things
That’s the gap Carly fills. It connects to Linear natively — no endpoint to configure, no OAuth cache to clear — plus around 260 other apps and anything with a public API through your own key. The difference is the trigger: Carly’s workflows fire on events and schedules, so Linear work happens with nobody’s terminal open.
- When an issue moves to Blocked → post the reason to the team channel and comment the last three commits touching that file.
- When a P0 lands in triage → page the on-call, open the incident doc, and link it back on the issue.
- Every Friday → summarize the cycle’s slipped issues and send it to the team lead.
The non-AI steps — routing, matching, moving data between apps — are free and unlimited, the Zapier-style backbone. The AI steps start at $35/month. You describe the outcome in plain language and Carly wires up Linear and whatever sits downstream of it.
FAQ
Does Linear have an official MCP server?
Yes. Linear hosts and maintains a remote MCP server at https://mcp.linear.app/mcp, launched May 1, 2025 and expanded in February 2026. It’s not a community project — Linear runs it, and it’s a first-party connector in Claude.
What can the Linear MCP server do? Linear documents it as finding, creating, and updating issues, projects, and comments, with initiatives, initiative updates, project milestones, project updates, and project labels added in February 2026. It also loads images and Linear resources by URL.
Do I need a paid Linear plan for MCP? No separate purchase. You’re authorizing an MCP client against your existing Linear workspace under your own permissions, so whatever you can see and edit is what the server can. Workspace admins can restrict access with allowlists and workspace-level MCP permissions.
Why did my Linear MCP connection stop working?
The most common cause in 2026 is a config still pointing at the old https://mcp.linear.app/sse endpoint, which Linear began retiring in February in favor of https://mcp.linear.app/mcp. If the URL is right, clear the auth cache with rm -rf ~/.mcp-auth and reconnect.
Can the Linear MCP server notify me when an issue changes? No. MCP is request/response — the server answers when a client asks and has no way to push. For “when an issue moves to X, do Y,” you need something with event triggers, like a workflow tool, not an MCP server.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
