MCP Reference

Win & Scale is MCP-first. An agent runs the whole product through meta_ tools, with the schema of every tool in this reference.

Win & Scale is built for agents. The primary way to drive it is the meta_* MCP tools: the same tools the REST API runs, behind one endpoint.

https://winandscale.com/api/mcp

An agent can read what an ad account has bought, review and correct how its creatives are tagged, research customers, brief and commit a Test, and launch the next ads paused. A person turns spend on.

This reference lists every tool with its input schema, grouped by silo. For connecting a client see Connecting. For the safety model see Guardrails.

How the tools behave

  • 84 tools, all prefixed meta_. They return { ok: true, ... } or { ok: false, error }. Read the error: it usually says what to do next.
  • No account id. A workspace is one Meta ad account, so no tool takes a projectId. Call meta_get_workspace first to see which account every call hits.
  • Server instructions are delivered to every client at the MCP initialize handshake. They cover orientation, the evidence-first rule, the golden paths and the rules below.
  • One payload. Where a tool has a REST endpoint, both return the same result.

Read vs write

Each tool page carries a badge.

  • read Reads. They carry readOnlyHint, and a read-scoped key is offered only these tools.
  • write Changes stored data, spends, or changes delivery. Some are marked destructive: anything that changes live delivery, loses data, or overwrites a record in place with no version kept.

40 tools are reads and 44 are writes. A write is not necessarily money-moving: syncing, tagging and saving a brief are writes because they change stored data. The money-moving ones are listed under Live changes and in Guardrails.

Tool silos

From zero to a first launch

A typical path for an agent:

  1. meta_get_workspace, then meta_get_evidence_health. Quote its warnings.
  2. meta_get_account_history (pull it first if it is empty), then meta_get_history_winners for verdicts, Traits and Next tests.
  3. meta_save_briefs, then meta_create_test to commit.
  4. After the creative is made, meta_launch_ads with the testId. Everything is paused.
  5. The person reviews. Only on their word, meta_activate.

The step-by-step versions are in the guides.

On this page