How do AI agents get social data?
Agents should call documented APIs, not invent endpoints from memory. On Social Fetch that means MCP tools, POST /v1/ask for routing, Agent Skills for standing instructions, or typed REST with an API key, all returning the same envelope.
Three integration paths
MCP: OAuth into https://api.socialfetch.dev/mcp and call tools mapped from the public catalog.
Ask: natural-language routing when the agent does not know the path yet. Pin the routed operation afterward.
REST: API key + OpenAPI. Best for fixed workflows. Catalog: 17 platforms, 137 ops.
Agent hygiene
Read /llms.txt and /agents.txt. Install the Skills pack so the model is told not to invent routes. Log meta.requestId. Respect credits and 402.
Do not put an API key in a public MCP config file. Hosted MCP signs in with OAuth; REST keys stay in server env.
When agents should not scrape
Production agents that hit HTML selectors inherit every platform break as an incident. Prefer a maintained API so metering, errors, and requestId stay in one place.
Free try paths: /tools/ask and other helpers under /tools. Signup credits cover first API calls. For standing instructions, install the Skills pack under /docs/integrations/skills.
FAQ
Should agents scrape HTML themselves?
Usually no for production. Prefer a maintained API so breakages and metering stay in one place.
Where is MCP documented?
/docs/integrations/mcp and /glossary/mcp-for-social-data.
Is there a free way to try?
/tools/ask and other free tools under /tools. Signup credits cover first API calls.