Developers · MCP
Native access for agents.
Connect an agent once; it can query every object your workspace follows.
Setup
Add to Claude Code or Cursor
- 1Copy the config block below.
- 2Add it to your MCP client — Claude Code, Cursor, or any MCP-compatible agent.
- 3Drop in your API key. The agent can now query every object your workspace follows.
add to Claude Code · mcp.json
{
"mcpServers": {
"recommend": {
"url": "https://mcp.recommend.co",
"apiKey": "rcmd_..."
}
}
}Tools
What the agent can call
| Tool | Description | Params |
|---|---|---|
| list_followed_objects() | The workspace watchlist. | — |
| get_object_state(id) | Current state of a followed object. | id: string |
| search_objects(query) | Find / resolve an object. | query: string |
| query_state(object, lens) | State for any object on demand. | object: string, lens: string |
| get_trajectory(id) | State over time. | id: string |
Example
Prompt → tool call → state
The agent asks a question, calls a tool, and gets a calibrated state back to act on.
One connection = the whole workspace, scoped by the key.
agent flow
① agent prompt
"Is facial SPF demand worth briefing creators on this week?"
② tool call
get_object_state({ id: "demand:facial-spf" })
③ state returned
{ "state": 73, "verdict": "watch",
"move": "Brief 2 creators before the UV peak" }
④ agent acts
"Demand is at 73 (watch) and rising — yes. Recommended move:
brief 2 creators before the UV peak."Developers
Coming soon
We're putting the finishing touches on this section. Try the live playground in the meantime.
Try the playground