Documentation
EspañolAgent overview
Give an agent useful Mico context without exposing storage internals or unbounded transcripts.
Mico gives agents three read-only ways to access meeting context. Choose based on where the agent runs.
| Transport | Choose it when | Output |
|---|---|---|
| Remote MCP | The agent speaks MCP over HTTP, on any machine | Structured tool results; requires Flow and a personal access token |
| CLI | The agent can run local shell commands | JSON with --json, or Markdown for people |
| REST API | Your own code needs meeting data over HTTPS | JSON with cursor pagination; requires Flow and a personal access token |
Recommended agent flow
- List recent meetings or search by a short title fragment.
- Use a returned meeting ID. Never infer or invent one.
- Fetch the meeting detail before requesting transcript text.
- Fetch recurring history only when earlier meetings are relevant.
- Read transcript pages only until the task has enough evidence.
Prefer MCP when it is connected because each tool carries its input schema. Use the CLI when the agent only has shell access. Both support the same bounded transcript workflow.
Safety boundary
Agents should use a Mico interface, not SQLite. The supported interfaces apply Mico's rules for canonical notes, generated summaries, excluded participants, recurring meetings, and transcript text.
The CLI, the MCP server, and the API are read-only. An agent cannot update a note, action item, or meeting through them. CLI export can create a separate file; replacing a file requires --force.
CLI access stays on your computer. The remote MCP server and the API read the meetings you sync to Mico's servers, without keeping another copy, and only for a client holding one of your personal access tokens.
Use JSON output and resolve meeting IDs deliberately.
Connect the remote server and use bounded transcript pages.
For REST access from your own code, see the API reference.