2026-08-30

MCP for project management

Model Context Protocol is how a coding agent calls tools. It is not a backlog. People search “MCP project management” because they wired Claude or Cursor to Linear or Jira and still cannot say what shipped.

Linear has a first-party MCP. Jira has one. Both let an agent create an issue and move a status. That is useful. It is also the shallow end.

Creating a ticket is easy. Finishing a story with acceptance criteria, structured progress, and a human (or expressed) review gate is the job.

Be fair about the field

Linear’s GraphQL and MCP are good. The data model is clean. Agents waste fewer tokens than they do on Jira’s document format. If Linear is already the system of record and Linear Agent is the worker you want, use them.

Jira remains the enterprise process machine. Rovo and MCP exist. The model is heavier. We will not pretend otherwise.

Beck is not “we invented MCP.” We expose a queue an agent can actually finish.

What we install

npx @becklabs/beck-mcp-server

BECK_API_TOKEN=beck_…

BECK_API_URL=https://beck.bot

Tokens work on Free. Name the token after the bot. Point the agent at https://beck.bot/skill.md so it does not invent the process.

Tools cover workspace and documents, plan_it, recommend, start, log, finish, accept, reject, and minting worker tokens. GitHub is link_epic_github: a branch name and a PR URL. Not an app.

Step-by-step is in Connect an agent. The product page is MCP and API.

Lifecycle is the difference

Recommend the next allowed story. Start it. Log what was attempted, what succeeded, what failed. Finish. Someone accepts or rejects on the same object. That is project management the agent can operate — not a human UI with a sidecar.

If you already live in Linear, keep Linear. If you want one list for a small team and the agents you already run, the protocol is the easy part. The queue is the product.

See Plans.