Micro

Developers

Use your Micro assistant from your own code. Same account, conversations and usage allowance.

Ask your assistant

Create an API token, leaving Agents and Allow actions selected. Set it as MICRO_TOKEN in your environment, then run:

curl 'https://micro.mu/api/v1/agent/ask' \
  -H "Authorization: Bearer $MICRO_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"prompt":"Help me plan my week"}'

The response contains data.text (the reply) and data.thread (the saved conversation ID). To continue, send another request to the same endpoint with:

{"prompt":"Focus on Monday", "thread":"THREAD_ID"}

Replace THREAD_ID with the returned ID. Omit it to start a new conversation. Keep your token on your server, outside browser code and source control.

Requests use your account’s included allowance, then prepaid credit. Billing · Usage costs.

MCP

Add a remote server using HTTP with this URL:

https://micro.mu/mcp

Set its authorization header to Authorization: Bearer YOUR_TOKEN, using the same API token. The client can discover tools and call agent_ask with a prompt. Tool reference.

API and tool reference

All calls use POST with the same authorization and JSON headers as the example above. Enable the corresponding capability on your token. Actions also require Allow actions.

Successful responses wrap the result in {"data": ...}. Errors return {"error":{"code": ..., "message": ...}}. A 401 means the credential is missing or invalid; 403 means access is not allowed; 402 means there is not enough credit.

For background work, submit a job once, then poll its ID with Work get. Do not automatically resubmit after a lost response: it could run twice.

agent_list

List the agents you can ask.

POST /api/v1/agent/list

agent_ask

Ask an agent. Pass the returned thread ID to continue the saved conversation.

POST /api/v1/agent/ask

prompt — Your instruction, up to 8000 bytes.

agent — Agent name; defaults to Micro.

thread — An owned conversation ID to continue.

work_submit

Start background work. Returns a durable ID; use work_get for the result. Do not blindly retry submissions.

POST /api/v1/work/submit

prompt — The goal, up to 8000 bytes.

agent — Agent name; defaults to Micro.

thread — Optional owned conversation for context and result delivery.

work_get

Read work status, result and execution steps.

POST /api/v1/work/get

id — An owned work ID.

work_list

List delegated work: open jobs first, then recent completions. Personal todo items are excluded.

POST /api/v1/work/list

status — Optional todo, doing, done, failed or blocked filter.

offset — Number of matching jobs to skip.

limit — Page size, 1–100; default 20.

work_retry

Explicitly retry reviewed failed or blocked work. Earlier side effects may be repeated.

POST /api/v1/work/retry

id — An owned work ID.

inbox_list

List recent conversations across channels. Reading this list does not mark messages read.

POST /api/v1/inbox/list

limit — Maximum items, 1–100; default 20.

inbox_read

Read the latest messages of a conversation without changing its unread state. Use agent_ask with its thread ID to ask about it.

POST /api/v1/inbox/read

id — An owned conversation ID.

limit — Maximum items, 1–100; default 20.

inbox_mark_read

Mark a conversation read.

POST /api/v1/inbox/mark_read

id — An owned conversation ID.

inbox_mark_unread

Mark a conversation unread.

POST /api/v1/inbox/mark_unread

id — An owned conversation ID.

Micro