Micro “restarted” with an idea around tools for agents: create one MCP endpoint for search, news, weather and other services which become capabilities your agent can call using just one API token to do some work. Now there’s more.
Micro let’s you create an agent with a basic system prompt, ask a question and assign work to run in the background. This can be done via CLI or HTTP. You don’t need to configure a model or run a server to use the hosted service.
Login to start
Build the current CLI from the Mu repository:
git clone https://github.com/micro/mu.git
cd mu
go build -o mu .
./mu login https://micro.mu
Login takes you to the token page. Choose Agents / Account, enable Agents and Background jobs, and enable Allow actions. Add Inbox if you also want to read conversations through the API.
Then ask something to test the default agent:
./mu ask "Compare SQLite and PostgreSQL for a small personal server"
The conversation is saved in your account. Add --raw to receive JSON with the answer and its thread ID. Continue it with ./mu ask --thread THREAD_ID "What would you choose for a single-user app?".
Give it work
Sometimes you want to hand over a task and come back to the result:
./mu work submit --prompt "Compare SQLite and PostgreSQL for a small personal server. Cite sources and recommend one."
./mu work get --id WORK_ID
Submission returns an id. Use it to check progress. The response contains a work object with the status, result, steps and attempts. The same task is visible on the web in Work.
Queued work has status todo; running work is doing; completed work is done. A failed, blocked or canceled task needs review. Check periodically rather than holding a connection open. If a submission response is lost, inspect your work list before submitting again: the first request may have succeeded.
Create a focused agent
For work you do repeatedly, give an agent its own instructions and tools:
./mu agent create researcher \
--prompt "Research questions using sources. Cite URLs and distinguish facts from uncertainty." \
--tools web,news
The tool scope uses service names. This agent can use Web and News; it doesn’t need access to Mail or your other services to do that job. Creating it doesn’t create another token.
Use the agent name returned by creation when assigning work:
./mu work submit --agent researcher \
--prompt "Compare SQLite and PostgreSQL for a small personal server"
You can also talk to it with ./mu ask --agent researcher. Your plan’s agent limit and the normal credit charges apply.
Call it from your application
The CLI uses ordinary HTTP. With your token in MU_TOKEN, the same work submission is:
curl https://micro.mu/work \
-H "Authorization: Bearer $MU_TOKEN" \
-H 'Content-Type: application/json' \
-H 'Accept: application/json' \
-d '{"agent":"researcher","prompt":"Compare SQLite and PostgreSQL for a small personal server"}'
Use POST /agent to ask, POST /agents to create an agent, and GET /work?id=WORK_ID to retrieve a task. Pass an owned conversation’s thread ID when submitting work if you want its context included and the result delivered back there. Inbox is where you can read those conversations and updates.
The developer guide has the request bodies, token settings and examples.
Hosted or your own server
Micro is the hosted service. Mu is the open source runtime behind it. To use your own configured Mu server, run ./mu login https://your-server.example. The commands and HTTP resources stay the same. Self-hosting requires your own model and service configuration.
If you already run an agent and only need capabilities, MCP tools and the services HTTP API remain available. x402 provides wallet-paid access to public services through m3o.com, the machine-readable endpoint. Hosted agent work uses your Micro account and credits.
Use the tools directly, or give an agent the work. Start with the developer guide.