Point Claude at your GTM data

The fastest-growing class of AstroFabric callers is not backend services. It is AI assistants: Claude Code in a terminal, Cursor in an editor, custom agents inside our customers' products. They speak the Model Context Protocol, so the platform does too, natively.

{
  "mcpServers": {
    "astrofabric": {
      "url": "https://www.astrofabric.ai/api/mcp",
      "headers": { "Authorization": "Bearer ek_live_..." }
    }
  }
}

That one entry gives the assistant the full data catalog with input schemas: company and person lookups, waterfall enrichment, verification, signal feeds, list and audience operations, and the AI agents as outcome-shaped tools. Ask "which of these 200 accounts started hiring for RevOps this month, and who owns it at each?" and the model can pull the hiring signals, enrich the rows, find and verify the people and answer with the evidence attached, in one conversation - with the CRM write parked for your approval.

Same key, same rules

The MCP surface authenticates with the same API keys as REST, honors the same scopes, bumps the same meters and hits the same credit ledger. There is no side door where an enthusiastic agent can spend outside the envelope; a credit stop reaches the model as a readable error it can explain back to you, and every external write it attempts lands in the same approval queue and the same audit log.

If you are building an assistant rather than using one, the same server works as a remote toolbox for any MCP-capable framework. Mint a key with narrow scopes and hand your agent exactly the capabilities you want it to have.