Linear alternative for AI agents: Weft for ChatGPT and Claude

If AI agents such as Claude Code, Codex, Cursor or ChatGPT now do part of your work, a good Linear alternative is a board built around the agent's loop: the agent claims a task with a lease, asks you a question when it needs a decision, and finishes with evidence you can check. Weft is one such board, reached over MCP, with setup guides for 33 AI clients. If you are an engineering team that lives in cycles, triage and pull requests, Linear is excellent and you should probably stay on it.

We make Weft. Facts about Linear come from its official pages, checked on 3 November 2026. Facts about the other tools in the second table come from their official pages, checked on 27 October 2026.

Why do people look for a Linear alternative?

Most people do not leave Linear because it is slow or badly made. It is fast, and it is very good at what it was built for: software teams planning work in cycles, sorting incoming requests in triage, and linking issues to GitHub pull requests.

The reasons to look elsewhere are narrower, and they fall into two groups.

• The agent became a worker, not a helper. When Claude Code or Codex picks up a task, runs for twenty minutes and comes back with a pull request, the board needs to know who holds the task right now, what happens if the agent dies mid-run, and what proof came back. An issue tracker records the issue; it was not designed around that loop.
• The tracker is heavier than the work. A solo founder or a three-person team with one project often does not need teams, cycles, triage rules and an engineering workflow. They need four columns and an AI that can read and update them.

If neither describes you, this article will mostly confirm that Linear is the right tool.

Does Linear work with ChatGPT, Claude, Cursor and Codex?

Yes, in several ways, and it is worth being precise.

• Hosted MCP server. Linear runs one at https://mcp.linear.app/mcp over Streamable HTTP, with a read-only variant at /mcp/readonly. Sign-in is OAuth 2.1 with dynamic client registration, or a bearer OAuth token or Linear API key. Its docs list Claude (claude.ai, Desktop, Claude Code), Codex, Cursor, VS Code, Windsurf, Zed and others, and say the server can find, create and update issues, projects and comments (Linear MCP docs).
• Claude. Linear is in Claude's connector directory, made by Linear and marked Anthropic verified (claude.com/connectors/linear).
• ChatGPT. Linear's ChatGPT page describes a connector for deep research that helps you "explore and reference your team's work", for Plus and Pro users (excluding EEA, Switzerland and the UK) and for Team, Enterprise and Edu (linear.app/integrations/chatgpt). The page does not say it can write. ChatGPT is not mentioned in the MCP docs.
• Delegating to agents inside Linear. You can assign an issue to an agent, and agents "are not counted as billable seats" (Agents in Linear). The agents directory lists Codex, Cursor, GitHub Copilot, Devin and many others (linear.app/integrations/agents).

So if your agents are coding agents and your team works in Linear already, the connection exists and is well supported. The question is not whether an agent can reach Linear. It is what the board does while the agent is working.

What does an agent need from a board that an issue tracker does not give it?

Three things, in our view. Linear's MCP and agent docs, as of today, do not describe any of them as MCP operations. That does not mean Linear cannot be made to do something similar with conventions; it means the board does not enforce it.

• A claim with a lease. Two agents, or the same agent in two windows, should not start the same task. And if an agent crashes, the task should not stay "In Progress" forever. Linear's MCP page does not describe claiming or leases.
• A question to the person, from inside the run. When the agent hits something only you can decide, it should be able to ask, keep the task, and get the answer later. The Agents in Linear page does not say whether agents can ask people questions; it describes people @mentioning agents.
• Evidence on completion. "Done" from an agent is a claim. A useful finish says what was done, what was not, what is unknown, and links something you can check. Linear's GitHub integration does move a linked issue to Done when a PR merges (Linear GitHub docs), which is real evidence for code. For work that is not a pull request, there is no equivalent field.

How does Weft handle the agent's loop?

Weft is a Scrumban board (Backlog, Todo, Doing, Done by default, with optional sprints and projects) exposed through 32 MCP tools at https://letsweft.com/api/mcp. Your AI client signs in with OAuth, so there is no key to copy; personal access tokens exist for clients without OAuth. Weft issues refresh tokens, so the client renews the sign-in itself.

The loop is four calls:

start_task({ id })
  → runId, leaseToken          // 20-minute lease; a second agent gets already_claimed

report_progress({ runId, leaseToken })
  → lease renewed              // silence lets the lease lapse and frees the task

request_input({ id, runId, leaseToken, question, options })
  → question in your Inbox     // claim kept; push to your phone

complete_task({ id, runId, leaseToken, outcome,
                artifacts, notDone, unknowns, nextStep })

• Claim. start_task gives a 20-minute lease. report_progress renews it. If the agent goes quiet, the lease lapses and the task is free for someone else. A second agent that tries the same task gets already_claimed.
• Ask. request_input puts a question, with options, in the person's Inbox and sends a push to the mobile app. The agent keeps its claim and picks up the answer on its next call.
• Finish with evidence. complete_task records the outcome, artifacts (a URL, a commit, a file hash), what was not done, what is unknown, and the next step. By default the task goes straight to Done. It stops in your Inbox for Accept / Rework / Later / Drop only when the task needs review or a verification check fails.
• A limit on you, not on the agent. If 5 items are waiting on one person (10 for a team, or 45 minutes of review), start_task refuses with attention_queue_full. The agent is told to finish or clarify instead of starting more.

Each OAuth client and each access token is its own actor, and the board shows the client's registered name, so you can see which AI did what. Each connection can be given full, write or read rights, and you can disconnect it at any time. More on the loop: the claim loop for AI agent tasks and several AI agents on one board.

How do Linear and Weft compare side by side?

Sources for the Linear column: pricing, MCP docs, ChatGPT page, agents docs, GitHub docs and mobile page, checked 3 November 2026. A wider table is on our MCP task managers page, and the short version is at Weft vs Linear.

Is an issue tracker too heavy for a solo founder or a small team?

Sometimes. Linear's free plan is generous for a team: unlimited members, 2 teams and 250 issues. But the product's shape is an engineering organisation: teams, cycles, triage, projects and milestones. If you are one person with an AI client open all day, much of that structure is set-up you will not use.

Weft's shape is smaller on purpose: four columns, optional sprints, lightweight projects. Most of the time you do not open it to type; you tell ChatGPT or Claude "put these five things on the board" and look at the result on your phone. If that sounds closer to how you work, see AI task management for solo founders.

For small teams, the attention limit matters more than the column count. With three people and four agents, the bottleneck is not who can start work; it is how much finished work is waiting for someone to look at it. Weft refuses new starts when that queue is full.

What are the other Linear alternatives for AI agents?

A few other tools have hosted MCP servers. These facts were checked on their official pages on 27 October 2026.

None of these pages describe a claim with a lease or an agent-to-person question as MCP operations either. Sister articles: Asana alternative, Todoist alternative, ClickUp alternative and Motion alternative.

When should you stay on Linear?

• Your work is mostly code and pull requests. Linear's GitHub integration moves linked issues to In Progress when a PR opens and to Done when it merges. Weft has no GitHub integration.
• You run cycles and a triage inbox. Linear's cycles carry unfinished issues over automatically, and Triage is "a special inbox for your team". Weft has optional sprints and no triage.
• You want Claude's directory, not a custom connector. Linear is listed and Anthropic verified. Weft is added in Claude as a custom connector.
• You want agents to be seat-free. Linear says agents are not billable seats, and its agents directory is large.
• Several teams, one workspace. Linear's Business plan has unlimited teams. Weft has one active sprint per workspace.

What does Weft not do?

To save you a trial: Weft has no GitHub or Slack integration, no public REST API or webhooks, no CLI, no desktop app, no Gantt or timelines, and no calendar time-blocking. It has no AI of its own and no AI credits; the thinking happens in the AI client you already use. Restoring a task from the trash is a Pro feature.

How much do Linear and Weft cost?

Linear: Free ($0, 2 teams, 250 issues), Basic $10 per user a month billed yearly, Business $16 per user a month billed yearly, Enterprise custom (linear.app/pricing). The page does not show a monthly-billed price for Basic.

Weft: Free is 50 tasks and 200 AI calls a month. Pro is $7 per seat a month, or $60 per seat a year. Details on letsweft.com/pricing.

How do I move from Linear to Weft?

1. In Linear, export your issues as CSV. A workspace admin can export everything from Settings → Administration → Import Export; anyone can export a filtered view from the command menu (members up to 250 issues at a time, admins up to 2,000). See Linear's export docs.
2. Create a Weft account at letsweft.com. The board starts with Backlog, Todo, Doing and Done.
3. Import the CSV in Weft. Linear CSV import is free on every plan; the step-by-step guide is at letsweft.com/migrate-from/linear. On Free, the 50-task limit applies, so archive what you no longer need before you import.
4. Connect your AI client: ChatGPT (the app, every plan), Claude Code (claude mcp add --transport http weft https://letsweft.com/api/mcp), Cursor (one click) or Codex (config, then codex mcp login weft). Sign in with OAuth when the browser opens.
5. Ask the agent to call get_my_work, pick a task, and work it through start_task → report_progress → complete_task. Check the receipt on the card.
6. Install the iOS or Android app so questions from request_input reach you as a push.

You do not have to move everything. You can keep engineering issues in Linear and put the work your agents do across code, writing and research on a Weft board. Setup for every client is in the docs; what MCP itself is is explained at modelcontextprotocol.io.

Linear already has an MCP server. Why would I need anything else?

You may not. Linear's MCP server lets an agent find, create and update issues, projects and comments. What its docs do not describe is a claim with a lease, a question from the agent to the person, or a structured receipt on completion. If your agents run long, in parallel, or on work that is not a pull request, those are the parts Weft adds.

Is Weft in Claude's connector directory like Linear?

No. In claude.ai and Claude Desktop you add Weft as a custom connector with the URL https://letsweft.com/api/mcp. In Claude Code it is one command: claude mcp add --transport http weft https://letsweft.com/api/mcp. Linear is in the directory and Anthropic verified.

Does Weft work with ChatGPT on the free plan?

Yes. Weft is a published app in ChatGPT and works on every ChatGPT plan, including Free. See letsweft.com/chatgpt.

What happens if my agent crashes in the middle of a task?

The lease from start_task lasts 20 minutes and is renewed by report_progress. If the agent stops reporting, the lease lapses and the task is free for another agent or a person. It does not stay in Doing under a dead run. More in what did my AI agent do.

Will every agent finish land in my review queue?

No. By default a completed task goes straight to Done with its receipt. It stops in your Inbox for Accept / Rework / Later / Drop only when the task needs review or a verification check fails.

Can I import from Linear on the free plan?

Yes. Linear CSV import works on every Weft plan. The Free plan holds 50 tasks, so a large Linear workspace will need Pro or some archiving first. Guide: letsweft.com/migrate-from/linear.