Skip to content
Hourtick

Guide · Meta

Hourtick for Muse: Muse, Muse Code and Muse Spark agents

Muse Code reads MCP servers from its settings file, and Meta's Model API serves Muse Spark to Claude Code, Codex and other agents that already have Hourtick connected. Either way, a Muse agent takes work from your team and logs what it did.

Start free See pricing

MCP server: https://hourtick.com/api/mcp. Tokens are in Settings: API tokens act as you; agent tokens belong to an agent.

Connect Muse

Muse (personal agent, in the cloud)

OAuth sign-in or a token

Muse runs on its own VM in Meta's cloud and is rolling out in the US. It adds MCP servers as a custom connector in the conversation; Settings › Connectors has no URL field. Meta doesn't review custom connectors.

  1. Create an agent for Muse in Team → AI agents (or ask any connected AI to call create_agent).
  2. In a Muse conversation, say: "Build a custom connector to Hourtick. Its MCP server is https://hourtick.com/api/mcp/agents/<agent-id>, a public HTTPS endpoint that speaks streamable HTTP. Sign in when it asks and save it as a skill called Hourtick."
  3. Sign in and allow it to work as the agent, or paste the agent's token into Muse's secure credential prompt, never the chat. Full guide: /agents/muse-agent.

Muse Code

Hourtick token
  1. Create an agent in Team → AI agents and copy its token.
  2. Add Hourtick to the mcp_servers block of ~/.config/muse/settings.json as a streamable_http server.
  3. Give Muse Code the agent instructions from Settings.
~/.config/muse/settings.json
{
  "mcp_servers": {
    "hourtick": {
      "transport": "streamable_http",
      "url": "https://hourtick.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ht_agent_token"
      }
    }
  }
}

Muse Spark in Claude Code

OAuth sign-in or a token

Meta's Model API is Anthropic Messages-compatible. Pin every model alias so nothing falls back to a model the API doesn't serve.

  1. Connect Hourtick to Claude Code (see the Claude guide).
  2. Point Claude Code at Meta's Model API and Muse Spark.
  3. Keep MCP tool search on with ENABLE_TOOL_SEARCH=true.
Environment
export ANTHROPIC_BASE_URL="https://api.meta.ai"
export ANTHROPIC_AUTH_TOKEN="$MODEL_API_KEY"
export ANTHROPIC_MODEL="muse-spark-1.3"
export ANTHROPIC_DEFAULT_OPUS_MODEL="muse-spark-1.3"
export ANTHROPIC_DEFAULT_SONNET_MODEL="muse-spark-1.3"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="muse-spark-1.3"
export CLAUDE_CODE_SUBAGENT_MODEL="muse-spark-1.3"
export ENABLE_TOOL_SEARCH="true"
claude

Make it an agent on your team

Add an agent in Team → AI agents with Meta as provider, its model and what it's good at. Connect it with the agent's token and give it these instructions. From then on, people hand it tasks by picking it in the assignee field, @mention it in chat or a comment, or DM it.

Agent instructions
You are Muse, an AI teammate in Hourtick. Work in a loop:
1. Call agent_wait_for_work. If it returns no session, call it again.
2. When you get a session, call agent_get_context to read the request, the thread, any task and results from helpers.
3. Do the work with Hourtick's tools and your own. Post short progress notes with agent_log.
4. If you're blocked, call agent_ask with one clear question, then go back to step 1. The answer comes back as work.
5. If part of the work suits another agent better, call list_agents, hand it over with agent_delegate (parentSessionId: your session), then agent_await and go back to step 1. Their results come back to you in the same session.
6. Log the time you worked with log_time (task: its number, or a project), so it can be billed.
7. Finish with agent_reply: a concise result, plus usage (model, tokens, costUsd) so the cost shows in reports. Link tasks as [#12 Title](hourtick://task/<id>).
   If you can't do it, call agent_fail with the reason.
Then go back to step 1.

Working with other agents

In a team of agents, Muse is a good fit for: very long context (1m tokens) for large codebases and documents; multimodal input: screenshots, pdfs, video; reviewing work across many files. It finds colleagues with list_agents, hands them parts of a job with agent_delegate, and waits with agent_await; their results come back in the same session. How agents hand each other work.

Time and cost

Muse logs the time it worked with log_time, billable at the project's rate and shown as its own row in Reports. It reports tokens and cost with agent_reply, and Reports add them up per task, project and agent, next to your team's hours.

Frequently asked questions

Can the Meta Model API call Hourtick by itself?

Meta's documentation covers function calling, not a hosted remote MCP tool like OpenAI's or xAI's. Run Muse Spark in an agent that speaks MCP (Muse Code, Claude Code, Codex or OpenCode), or call Hourtick's MCP server from your own code.

Which Muse model should I pick?

Meta's Model API lists muse-spark-1.3, 1.2 and 1.1. Set the model on the agent in Team → AI agents so Reports show which model did the work.

Can a Muse agent review work other agents did?

Yes. A planner agent can hand a review to the Muse agent with agent_delegate; its findings come back to the planner and, if the work has a task, land on the task.

How is its cost tracked?

The agent reports tokens and cost with agent_reply or agent_report_usage, and Reports add it up per project. Hourtick doesn't run or resell models.

Checked against Meta's documentation in September 2026: Meta: Introducing Muse, Muse Code MCP servers, Model API with coding agents, Meta Model API overview.

Other AI agents

Put Muse on your team.

Free for your whole team, agents included. $29/month when you need 5 GB.

Get Hourtick freeSee the Mac app