Linear and AI agents: what runs on its own and what waits for approval
The agent reads the backlog, files issues from the conversation and sums up the week. Here is the level of each action.
· Sources
Short answer
In agentes.chat, Linear connects in one click through Linear's official MCP server, with an OAuth sign-in. The agent reads issues, projects and comments without asking. Creating and updating issues or commenting runs on its own once you grant write access. Tools that delete ask for approval or stay off.
How do I connect Linear to agentes.chat?
Linear is in the one-click connector catalog. A workspace admin picks Linear and signs in with their account. agentes.chat uses Linear's official MCP server at mcp.linear.app, which signs in with OAuth 2.1 and is read-write by default[1].
The token is sealed on the server and never sent to a browser. Each agent then gets only what it needs: no access, read, or read and write, or tool by tool. An agent that only sums up the backlog can stay read-only.
What can the agent do on its own in Linear?
Linear describes the server's tools as ways to find, create and update issues, projects and comments[1]. In agentes.chat, that means:
- Without asking: search and read issues, projects and comments. Reading and investigating never ask.
- Once you grant write access: create issues, update status, assignee and description, and comment.
So when someone in the channel says “file an issue for this”, the agent files it with the conversation's context and replies with the link.
What needs approval, and what stays off?
When Linear connects, agentes.chat suggests a level for each tool, and an admin reviews it and can change it. The server's own marks come first, then the verb in the tool's name, and a model only when nothing else decides:
| Kind of tool | How it is recognized | Suggested level |
|---|---|---|
| Only reads | The server marks it read-only, or the verb in its name is get, list, search… | Runs on its own |
| Creates or changes data | The verb in its name is create, update, add, set, assign… | Runs on its own once granted |
| May destroy data | The server marks it destructive | Asks for approval |
| Deletes or removes | The verb in its name is delete, remove… | Asks for approval, or stays off |
| Unclear (such as archive or move) | No rule decides | A model suggests; if it fails, it asks |
Sensitive tools can only ask or be switched off, whatever the suggestion says. The request is a card in the chat with the exact action. It runs once, exactly as approved, and expires after 24 hours. The agent's owner or an admin decides.
In routines, every write to a connected app waits for a person. A routine can read Linear and report freely, but filing an issue waits for someone to approve.
Before acting again on the same issue, the agent is told once what a teammate already did to it in the last 30 days. It goes ahead only with a reason.
What does it look like day to day?
Example · Acme (fictional company)
Acme's product week
Every Monday at 9 a.m., a routine wakes Acme's PM agent. It reads the current cycle in Linear and posts in #product what shipped, what slipped and what has no assignee. The routine only reads, so nothing asks for approval.
In the thread, someone asks: “file issues for the three bugs support mentioned”. Now a person is asking, not a routine, and the agent has write access. It creates the three issues in the right project, each with the relevant part of the conversation, and replies with the links.
agentes.chat also has its own task board, where an agent's task moves by itself as the agent works. If your team plans in Linear, your backlog can stay there.
How do I get started?
Linear is one of 58 one-click connectors in agentes.chat, all official MCP servers. See how the protocol works in MCP connectors and how routines run in routines and missions. agentes.chat is in early access and opens by invitation, in waves. Join the waitlist to bring your team's Linear to your agents.
Frequently asked questions
Do I need a Linear API key?
No. The one-click connection uses Linear's OAuth sign-in. A workspace admin signs in once, and the token is sealed on the server.
Can the agent create issues without asking?
Yes, if you grant it write access and someone asks in the conversation. In a routine, which runs without anyone asking, creating an issue waits for approval.
Can I keep an agent read-only?
Yes. Access is set per agent: none, read, or read and write, or tool by tool.
Can the agent delete issues?
Tools whose name has delete or remove as the verb, or that the server marks destructive, are sensitive: they can only ask for approval or be switched off.
Sources
Links checked on October 29, 2026. What this page says about agentes.chat describes the product on that date and may change during early access.
- 1Linear, MCP server.