Linear MCP

Connect Linear to Claude, ChatGPT, Cursor and Codex. Understand tools and permissions, then use worked triage, spec and cycle-report examples.

Published

Linear MCP

Linear MCP lets a compatible AI assistant retrieve your team's work and make issue updates from the conversation where you are already planning or coding. The useful change is less time finding tickets, copying context and retyping decisions. Linear hosts the server and exposes operations for issues, projects and comments. Linear's announcement

For a team already using Linear, start with a cycle summary, then add reviewed triage and spec-to-issue workflows. Claude, Cursor and Codex have documented MCP setup paths. Linear's documented ChatGPT connection is for deep research; that page does not promise issue editing. ChatGPT integration

The Server URL and Authentication

Use https://mcp.linear.app/mcp over Streamable HTTP, a web connection to the hosted server. Interactive setup uses OAuth 2.1 with dynamic client registration, which handles client registration and user authorization. Alternatively, send an OAuth token or Linear API key in Authorization: Bearer <yourtoken>. Linear MCP setup

Think of MCP as a shared service counter. Your assistant requests a named operation, such as finding an issue or saving a change. Linear's server carries out the operation allowed by the connection. Your prompt supplies the intent; it does not grant extra permissions.

ConnectionWhat Linear documents
https://mcp.linear.app/mcpRead/write access by default
https://mcp.linear.app/mcp/readonlyOnly read tools are exposed
/mcp with only the read OAuth scopeRead-only access; the token cannot reach write APIs
API key restricted to ReadAnother way to authenticate with read-only permissions

These are different ways to restrict a connection. A prompt saying “don't edit” is an instruction to the assistant; a read-only endpoint or credential restricts what it can do. Linear's access options

Architectural diagram connecting an AI client through OAuth to Linear MCP, then branching into read and write operations.
The standard connection can support reading and writing. The endpoint, granted permissions and client determine which operations are available.

Connect the Client Your Team Uses

Claude and Claude Code

In Claude's browser or desktop app, open Settings → Connectors and connect Linear. That is the path on Linear's Claude integration page.

For Claude Code:

  1. Run claude mcp add --transport http linear-server https://mcp.linear.app/mcp in your terminal.
  2. Open a Claude Code session.
  3. Run /mcp and complete the authentication flow.

Both the command and follow-up action come from Linear's Claude setup.

Cursor

Linear's main guide offers an install link and directs you to find Linear on Cursor's MCP tools page. Use the installation entry in Linear's Cursor section.

Its dedicated Cursor page also gives this manual route:

  1. Press Ctrl/Cmd + Shift + J to open Cursor Settings.
  2. Select MCP.
  3. Select Add new global MCP server.
  4. Add this configuration:
JSON
{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.linear.app/mcp"]
    }
  }
}

This uses mcp-remote, a bridge between the client and the hosted server. It is the configuration published on Linear's Cursor integration page.

Codex CLI and IDE Extension

Linear says the CLI and IDE extension share their MCP configuration.

  1. Run codex mcp add linear --url https://mcp.linear.app/mcp.
  2. Follow the Linear login prompt described in its setup guide.
  3. For the manual configuration route, Linear directs you to authenticate with codex mcp login linear after editing ~/.codex/config.toml.

There is a documentation discrepancy worth catching before you copy a configuration:

Linear pageFeature setting it prints under [features]
Main MCP guideexperimental_use_rmcp_client = true
Dedicated Codex integrationrmcp_client = true

The dedicated page's manual server entry is [mcp_servers.linear] with url = "https://mcp.linear.app/mcp". These settings are reproduced as published, but the conflicting feature flags should not be treated as a universal current prerequisite. Linear's Codex integration, main MCP guide

Use the add/login route above. OpenAI's current MCP guide documents HTTP server configuration and codex mcp login <server-name> without requiring either feature switch. Official OpenAI MCP documentation

ChatGPT

Follow the route Linear actually documents:

  1. Open ChatGPT's connector settings using the link on Linear's ChatGPT integration page.
  2. Connect Linear.
  3. In deep research, add Linear as a source.

Linear describes this connection as a way to reference issues, projects and discussions while preserving existing user-level permissions. It does not document issue creation or updates through this setup. Use it for research and summaries; use a documented write-capable MCP client for the editing examples below. This distinction is about that connection, rather than a blanket claim about every ChatGPT integration.

The Tools Linear Publicly Names

The useful distinction is between retrieving context and saving changes. These are the exact identifiers in Linear's published commonly used tools list, alongside their purpose. That list appears on its Slack integration page in the section about connecting the same Linear MCP server to Slackbot. Linear's tool list

Tool identifiersPurpose
list_issues, get_issueFind issues; documented filters include team, assignee, project, label, status, priority and cycle
save_issueCreate or update an issue, including its title, team, assignee, project, labels, priority and status
list_projects, get_projectRetrieve projects and details such as milestones and members
get_status_updatesRead project or initiative updates
list_usersFind workspace people for ownership and assignment

Linear's changelog also names list_cycles, list_issue_labels, list_comments, save_project and save_document. It documents cycle details, paginated labels and comments, project changes, and creating or reparenting documents under initiatives or cycles. This is a published inventory of relevant tools, not a claim that every client exposes an identical, fixed-size catalog. Cycle and label tools, comment tools, project and document tools

Older examples may call create_issue and update_issue separately. Linear combined them into save_issue. Keep that in mind when copying old prompts or building around tool names. Linear's tool change

What Access It Gets and What an Admin Controls

Treat the connection as an authorized route into workspace data. In Linear's OAuth terminology, read means “Read access for the user’s account” and write means “Write access for the user’s account.” The default user actor creates resources as the person who authorized the application. Linear OAuth permissions

For API keys, Linear's wording is “full access to the data your user can access,” with the option to restrict permissions and teams. That makes a restricted key materially different from a broadly authorized personal key. API-key permissions

The controls depend on how the connection authenticates:

ControlWho manages it and where
Third-party application approvalAn Enterprise workspace owner, or an admin in another paid workspace, enables approvals in Settings → Administration → Security. Approved and denied apps appear in Applications.
Member API-key creationAdmins use Settings → Administration → API → Member API keys. They can also view and revoke workspace API keys there.
Key permissions and team accessWhen creating a personal key in Settings → Account → Security & Access, permitted users can restrict permissions and teams.

Linear describes the purpose of application approvals as controlling “which applications can access the workspace and its data.” Members can request approval during installation. Key controls apply to API-key authentication; they are not a substitute for governing OAuth app connections. Application approvals, API administration

For supported Anthropic clients, Enterprise workspaces with Okta SAML can use enterprise-managed authorization. Linear says this “applies their existing Linear permissions.” Administrators manage access through Okta rather than requiring separate user authorization for each connection. Enterprise-managed authorization

For a status-only workflow, choose read-only access. For triage and issue creation, grant the required write access and make the proposed changes visible before applying them. That review step is a workflow recommendation, not a claim that every client enforces the same approval behavior.

Three Workflows Worth Using

Start with cycle reporting because it can work without writes. Triage is next when intake is repetitive. Spec conversion pays off when a lead is spending time translating settled decisions into tickets.

The examples below are fictional worked examples. They illustrate prompts and expected deliverables, not results from a signed-in test.

Triage New Issues Into a Reviewable Proposal

An engineering lead could ask the assistant to inspect incoming reports, retrieve similar existing issues and propose changes together. The payoff is less repeated searching and property editing, while the lead retains the decision about ownership and priority.

Example input: A new report, “Export button hangs,” says the browser stops responding during a filtered export. An existing issue, “Filtered export never finishes,” describes the same screen. Another report, “Invites never arrive,” has no delivery logs or reproduction steps.

Prompt:

Review the new issues in our team's Triage queue. Retrieve each issue and search existing issues for relevant matches. Propose labels, an assignee and priority using our written triage rules. Explain the evidence for each proposal. Treat similar reports as possible duplicates until reviewed. Show the current and proposed values before changing anything. Flag missing information instead of guessing.

A useful proposed result:

Incoming issueProposed treatmentEvidence or gap
Export button hangsFlag a possible duplicate and propose the existing export issue's owner for reviewBoth descriptions mention filtered exports; matching symptoms do not prove a shared cause
Invites never arriveKeep in triage and request delivery detailsThe report does not establish whether sending or receiving failed

After review, the lead could authorize just the agreed label and assignee changes through save_issue, leaving ambiguous reports untouched. The assistant should return the affected issue links and a record of the fields changed. The retrieval and save operations are in Linear's tool list.

Before building a custom triage system, compare it with Triage Intelligence. Linear already offers property suggestions and duplicate or relationship detection. Custom MCP orchestration makes more sense when the decision needs context or rules beyond the native workflow. Triage Intelligence

Turn a Spec Into Issues Without Inventing Requirements

A founder could move from an approved feature brief to draft issues, then create the reviewed work in Linear. The payoff is consistency: each ticket carries the requirement it came from, and missing decisions stay visible.

Example spec: “Workspace admins can invite members by email. Invitees accept through an email link. Admins can revoke pending invitations. Billing changes and bulk invitations are out of scope.”

Prompt:

Read this spec and inspect the existing project for overlapping work. Propose a parent issue and sub-issues with titles, descriptions, acceptance criteria and links back to the source requirement. Keep explicit exclusions in the parent description. List unanswered product questions. Do not invent owners, deadlines or dependencies. Show the proposal before creating issues.

Proposed issue breakdown:

Draft issueAcceptance criteria grounded in the spec
Invite members by emailA workspace admin can send an invitation to an email address
Accept an invitationThe recipient can accept through the emailed link
Revoke a pending invitationA workspace admin can revoke an invitation that has not been accepted

Open questions: What happens when an email already belongs to a member? Does an invitation expire? What should an invitee see after revocation? These stay questions until the product owner decides.

Once approved, the assistant could create the parent and sub-issues, return their URLs and identify any item it could not create. Sub-issue creation is a documented MCP capability. Linear's MCP changes

The assistant performs the transcription. The product owner still resolves the product decisions.

Write a Weekly Status From the Cycle

A lead could turn cycle work into a status draft with links supporting each claim. The payoff is a shorter path from scattered issue updates to a report someone can check.

A weekly report and a cycle summary can cover different periods. Specify the team, cycle and reporting window, so “completed in this cycle” does not silently become “completed this week.”

Example source material: “Invitation email” is marked Done during the reporting window. “Accept invitation” remains In Progress. A comment on “Revoke invitation” says implementation is waiting on a product decision about the revoked-link experience.

Prompt:

Find the current cycle for our team and identify its dates. Gather the relevant issues and comments, following all result pages. Draft a weekly update for the reporting window I provide. Separate work completed during that window, work still in progress, recorded blockers and decisions needed. Attach source issue URLs to every substantive claim. Distinguish an issue marked Done from evidence that a feature shipped. Say when completion timing is unclear. Do not change Linear or send the report.

Example draft:

Completed: Invitation email is marked Done within the reporting window. In progress: Invitation acceptance remains open. Decision needed: The revoked-link experience is blocking invitation revocation. Release status: The supplied issue evidence does not establish that the invitation flow has shipped.

In a real draft, each sentence should carry the corresponding issue URL. The example deliberately makes no completion percentage or release claim. Cycle, issue and comment retrieval provide the ingredients for this workflow. Cycle and issue tools, comment retrieval

Cycle issues split into Done and In progress channels, then pass through an Issue links review before becoming a Weekly update.
Keep completed and unfinished work separate, then check the supporting issue links before sharing the update.

Other Useful Jobs Once Those Workflows Are Stable

SituationWorkflow a team could useWhy it helps
A lead finishes standup with scattered notesMatch notes to existing issues, review proposed comments, then post the accepted updatesLess repeated typing and fewer updates lost in chat
An engineer takes over an unfamiliar projectRetrieve the project, related issues and recent updates; draft a handoff with unresolved questions and source linksLess time asking teammates to reconstruct context

Comment creation is part of the server's documented capability, while the project tools supply context for a handoff. Server capabilities, project tools

The Business Math: Count Review Time Too

The budget line affected is engineering coordination. Before MCP, a lead searches, copies, reconciles and writes. With an assistant connected, the lead still scopes the request, reviews the result and corrects mistakes.

Use this break-even calculation with your own costs:

Net value = (manual preparation time − assisted preparation, review and correction time) × loaded hourly cost − incremental client and operating costs.

A polished report that takes longer to verify has not saved the team time. Keep the same report scope when comparing the workflows, and count cleanup after incorrect edits. The useful target is less coordination work per accepted outcome.

For subscription choices, see Linear pricing. If the underlying tracker is still undecided, Linear vs Jira addresses that choice. Our best MCP servers guide covers adjacent connections.

What Is Worth Building Around It

A cycle status drafting service is the strongest starting opportunity. An engineering lead could pay for a consistent report that respects team boundaries, separates finished work from open work and includes issue evidence. The smallest useful version would save a report definition, retrieve the relevant cycle data and produce an editable draft. Scheduling, retries and delivery would be part of your application.

The catch is differentiation: a generic summary prompt is easy to reproduce. The potential product is the company's reporting rules, source traceability and review process. Confirm that teams repeatedly struggle with those details before building a standalone subscription.

A custom triage review queue is a narrower opportunity. A platform or support-engineering lead could pay to combine Linear issue context with their own routing policies and separately authorized support context. The smallest version would show proposed field changes, evidence and an approve-or-skip action, then apply accepted changes.

The catch is direct competition with Linear's native Triage Intelligence. Build only where the extra context or policy changes a decision the native system cannot handle adequately for that team. These are product hypotheses, not claimed customer outcomes.

Where the Connection Stops Helping

MCP gives the assistant access to operations; it does not settle ambiguous requirements or make incomplete issue data reliable. A ticket marked Done can support a status statement about that ticket. It cannot, by itself, prove a customer received the feature.

Likewise, authorizing Linear does not authorize every document, repository or support system mentioned in an issue. For a build that needs those sources, plan the separate access and retrieval paths.

For new setups, use /mcp. Linear labels /sse a deprecated fallback. The older jerhadf community server also explicitly says it is deprecated and no longer maintained, and directs users to Linear's official server. Current endpoint guidance

Do I need a Linear API key?

No. Use interactive OAuth, or a key if your client is configured for that alternative. Authentication options

Can ChatGPT update our issues through the setup above?

Linear's published ChatGPT setup describes deep-research access to issues, projects and discussions. It does not document writing issues through that connection. Choose a write-capable MCP setup for the triage and creation workflows. ChatGPT integration

Can I let an assistant read Linear without changing it?

Yes. Choose a read-only connection from the access table above for the reporting workflow. Read-only options

Can our admin decide which apps connect?

On paid plans, Linear provides third-party app approvals. Enterprise workspace owners, or admins in other paid workspaces, can enable them in the workspace's security settings. App approval controls

Your Monday move: have the engineering lead define the team, cycle and reporting window for a read-only status draft. Check its claims against the linked issues and compare the preparation effort with the current process. Expand to writes when the team has a clear review routine.

If you want these workflows built around your team's rules, we build AI automations.

Published
Category
Build
Related Articles
LangChain vs LangGraph

LangChain vs LangGraph

Choose LangChain or LangGraph for your app: current agent APIs, matching Python examples, migration tradeoffs and verified LangSmith deployment pricing.Oct 13, 2026Build
Agentic RAG: What It Is and How to Build It

Agentic RAG: What It Is and How to Build It

Agentic RAG versus classic retrieval: support, sales and finance examples, a small build, verified tool prices, and what to measure.Oct 13, 2026Build
Amp Code Review

Amp Code Review

Amp Code reviewed: current pricing, free tier, models, editor support and team features, plus when to choose Claude Code or Codex.Oct 12, 2026Build
Playwright MCP: Setup, Browser Tests, and Limits

Playwright MCP: Setup, Browser Tests, and Limits

Give Claude Code, Codex, Cursor, or VS Code a real browser. Exact setup, practical tests, a worked checkout failure, and the safety limits.Oct 12, 2026Build
Cursor Bugbot

Cursor Bugbot

What Cursor Bugbot checks, exact repository setup, usage-based billing, rules, Autofix, and current CodeRabbit and Greptile prices.Oct 12, 2026Build
Cline vs Claude Code 2026

Cline vs Claude Code 2026

Cline vs Claude Code: current plans, light and heavy monthly bills, model choice, editor and cloud support, and the limits that decide which to use.Oct 12, 2026Build
Best Claude Code Plugins 2026: Superpowers, Context7 and More

Best Claude Code Plugins 2026: Superpowers, Context7 and More

Compare nine Claude Code plugins for review, testing, docs, design, deployment and memory, with verified install commands, costs and safety checks.Oct 12, 2026Build
LiteLLM vs OpenRouter

LiteLLM vs OpenRouter

Compare LiteLLM and OpenRouter on keys, billing, fallbacks, budgets, caching and data handling, with a worked monthly bill and switching rules.Oct 12, 2026Build
Newsletter

One letter, every Sunday.Working systems, not hot takes.

Weekly. No spam. Unsubscribe anytime.