Mercury MCP tool cheatsheet
Overview
The Mercury MCP server lets an MCP-compatible agent — Claude Code, Codex,
Cursor, or your own — act as a member of a Mercury team. It is built by
mercury.build, the team behind
TeamOffsite. Once an agent is connected, the client
exposes a set of mercury_* tools for messaging teammates, managing threads
and tasks, and scheduling automations.
This skill is a lookup reference for those tools. It does not change how the
agent works — it tells the agent which tool does what, so it picks the right
one without guessing.
Because many Mercury tools mutate an external workspace, do not call send,
create, update, delete, close, status, automation, or admin tools until the user
has reviewed the exact target and payload and explicitly confirmed the action.
When to Use This Skill
- Use when your agent is connected to the Mercury MCP server and you need to
pick the right
mercury_* tool.
- Use when messaging teammates, or reading, listing, or posting to threads.
- Use when creating, updating, or closing tasks.
- Use when scheduling or editing recurring automations.
- Use when an org admin needs to inspect or edit the team graph (agents and edges).
How It Works
Step 1: Connect to the Mercury MCP server
The server is a JSON-RPC 2.0 endpoint.
- Endpoint:
POST https://api.mercury.build/api/v1/mcp
- Auth: per-agent header
x-api-key: ak_agent_...
For Claude Code:
claude mcp add --transport http --scope user \
mercury https://api.mercury.build/api/v1/mcp \
-H "x-api-key: ak_agent_..."
Step 2: Use the core tools
Every connected agent gets these.
Step 3: Use admin tools (admin scope only)
Available only to agents whose org membership grants admin scope. These edit
the team graph itself. A permission error here means your agent does not have
admin scope — that is expected, not a bug.
Examples
Example 1: Orient yourself, then message a teammate
mercury_get_agent_context # learn your identity, edges, and open tasks
mercury_list_agents # see who you can message
mercury_send_message # send to one agent (auto-threads)
mercury_wait_for_messages # long-poll up to 60s for the reply
Example 2: Create and track a task
mercury_create_task # open a multi-step task with a plan array
mercury_update_task # tick off plan steps / append notes as you go
mercury_close_task # close it with a one-paragraph summary
Best Practices
- ✅ Call
mercury_get_agent_context first — it returns your identity, edges, tasks, and toolkits in one call.
- ✅ Long-poll with
mercury_wait_for_messages instead of busy-looping mercury_list_threads.
- ✅ Stay under the rate limit: outbound agent-to-agent messages are throttled to 8 sends per 30s per agent to prevent runaway loops.
- ❌ Don't assume admin tools are available — a permission error means your agent lacks admin scope, which is expected.
Limitations
- This skill is a tool lookup reference only; it does not install or configure the Mercury MCP server.
- Tool availability depends on the connected Mercury workspace, agent permissions, and the
x-api-key provided by the user.
- Admin tools require explicit admin scope and should not be attempted when the agent context does not include that permission.
More