Can my AI agents use Mochlion, and what can they do?

Yes. Mochlion has an MCP server (the open protocol many AI tools support) and a token API. With a token, an AI tool can read a specification item by item, check how ready it is and what changed, ask a question back when something is unclear, and, if you add it as an expert, answer its own questions. It can't decide a disagreement, approve an item or sign off. Only people do that.

How does an agent connect?

A person on the project makes a token on the project's Connect tab. The token is shown once, and it acts with exactly that person's access: if their access changes or they leave the account, the token's access changes with it. Account owners can make tokens that reach every project on the account.

The project owner's team can see and revoke every token on the project. Each call is logged, and each token has a request limit.

What can a builder's AI tool do?

• Read the spec by trace ID: requirements, rules, data, acceptance checks and tests, one item at a time or by search.
• See where an item came from: the answers and decisions behind it, and what it links to.
• Download the build package for any version, and see what changed between versions after handoff.
• Check readiness and the handoff: how far along the spec is, and which version was handed over.
• Ask a question back. When the spec doesn't answer something, the tool asks about that trace ID. The question goes to the experts the item came from, and the tool reads the answer back once they've replied.

Through the API, a builder can also acknowledge a handoff.

What can an owner's agent do?

The same reading, with the owner's access: readiness across projects, the latest specification and package, and the change report. That makes it practical to keep a portfolio or governance tool up to date from Mochlion.

Can an AI agent answer questions as an expert?

Yes. Many agents already know the specifics of your systems. The project owner adds one as an expert (experts are unlimited and included) and gets its token once. That token reaches only that agent's own questions: it can read them, answer them, or say it doesn't know. The owner is responsible for the agent and what it submits.

What can't an agent do?

• Decide a disagreement between experts.
• Approve an item or confirm a change.
• Sign off the specification or declare it Build Ready.
• See other experts' questions, if it's an expert itself.

AI suggests; people decide. There is no tool or API call for deciding, approving or signing off.

Which AI tools work with it?

Any tool that can connect to an MCP server, or any software that can call a web API. Mochlion doesn't tie you to one AI vendor, and the build package is written to be read by any coding assistant or team.

How it works shows where agents fit in the overall process.