Skip to main content

​
How tools are named

Every tool is a verb and a noun: list_programs, create_brief, publish_brief, approve_payout, generate_ai_ad. Reads are free to call; writes change your workspace and are audited. Creating and publishing are separate steps, as in the app. A brief exists as a draft after create_brief and reaches creators after publish_brief. An agent that reads the tool descriptions follows this on its own.

​
Connect a smaller set of tools

The full catalog is more than 400 tools. Claude Code and Codex handle that; Cursor and some other clients warn or drop tools past a few dozen. Add a scope to the URL to connect a subset: Scopes combine: /mcp?profile=core&domains=ads is the core profile plus every ads tool. The rally_* helper tools below always come along.
A scope only narrows what the client sees. The key’s role still decides what the agent may do; a scope never grants more than the role allows.

​
Helper tools

Five tools exist only for agents and are present on every connection.

​
Resources and prompts

Alongside tools, the server offers:
  • Resources: Rally’s own documentation (rally://docs/...), the setup runbook and the program templates, so the agent can read how things work rather than guess.
  • Prompts: onboard_brand (set a workspace up from your website), weekly_review (health, exceptions, learnings and what to do about them), launch_campaign (a brief to published ads). Clients that support prompts list them as slash commands or templates.
The runbook is also a plain file at https://app.runrally.io/mcp/skill.md.

​
Limits

  • 600 calls a minute per key, of which 60 may be writes. Past that the agent gets a rate_limited error and should slow down.
  • Plan limits apply exactly as in the app. When one is reached the agent gets an error with an upgrade hint.