> ## Documentation Index > Fetch the complete documentation index at: https://docs.anarlog.so/llms.txt > Use this file to discover all available pages before exploring further. # Agent overview > Give an agent useful Anarlog context without exposing storage internals or unbounded transcripts. Anarlog offers two local, read-only agent transports. They share the same meeting model. | Transport | Choose it when | Output | | --------- | ------------------------------------------------------- | ------------------------------------------------ | | MCP | The agent supports a local stdio MCP server | Structured tool results and contextual resources | | CLI | The agent can run shell commands but MCP is unavailable | JSON with `--json`, or Markdown for people | ## Recommended agent flow 1. List recent meetings or search by a short title fragment. 2. Use a returned meeting ID. Never infer or invent one. 3. Fetch the meeting detail before requesting transcript text. 4. Fetch recurring history only when earlier meetings are relevant. 5. Read transcript pages only until the task has enough evidence. Prefer MCP when it is connected because tools carry their input schemas and results directly. The CLI provides the same bounded transcript workflow when an agent only has shell access. ## Safety boundary Agents should use the CLI or MCP server, not SQLite. Those interfaces apply Anarlog's rules for canonical notes, generated summaries, excluded participants, recurring meetings, and transcript text. All current Anarlog data operations are read-only. An agent cannot update a note, action item, or meeting through these interfaces. CLI export may create a separate file; replacing an existing file requires `--force`. Use JSON output and resolve meeting IDs deliberately. Configure the stdio server and use bounded transcript pages.