Social data your agent can call itself
One MCP endpoint and the API key you already have. The agent looks up what it needs when it needs it, instead of carrying the whole catalogue around.
Why your agent does not get 329 tools
The obvious way to put 329 endpoints on MCP is 329 tools. That is around 99,000 tokens of definitions loaded before the agent does anything, and it leaves the model picking between 329 similar-looking options. You pay for the context and the model still has a harder time choosing.
So the server gives it a small fixed set instead. One tool lists what is available, filtered by platform or search term, and comes back with each capability's parameters and credit cost. Another calls any of them by name. Two more are shortcuts for the things agents ask for most. The agent looks up what exists when it needs to, the same way you would check the docs.
Between those, an agent can reach 120 capabilities across 9 platforms. The tool list and the config to paste in are on the MCP page.
What the agent cannot reach, on purpose
209 of the 329 capabilities return pages of people who are not what you asked about: followers, following, reposters, repliers, and any search or feed where every result has a different author. None of those are on the agent surface. That is deliberate, and it is written into our privacy policy.
The reason is simple: pulling data on a few hundred other people should be something you decide to write, not something an agent does by itself while working on something else. All of those routes still work over REST, where a person chose to call them.
It cost us a feature to keep. An earlier version of the agent surface had a search tool that reached exactly this kind of capability. We removed the tool rather than soften the policy, and added a check that fails the build if any tool reaches one again. The full reasoning is in the privacy policy.
Why this works well for an agent
- Spending cannot run away. An agent stuck in a loop is exactly what runs up a bill. When your balance hits zero, requests are refused instead of billed, no plan has an overage rate, and cached, failed and declined calls cost nothing.
- Every response says what it cost. Your credit balance comes back with the data, so the agent can keep track of its own budget instead of finding out afterwards.
- It can look things up itself. The listing tool returns parameters and prices, so the agent can check what a call needs and what it costs before making it.
- The JSON is the platform's own. We do not rename anything, so a model working from a platform's public docs sees the field names it expects.
- Same key as everything else. One credential for REST and MCP, and you can revoke it per machine from the dashboard.
Point your client at https://api.socialpipe.dev/mcp with your key as a bearer token. Setup steps for Claude Code, Claude Desktop and Cursor are on the MCP page.
Pointing an agent at this
What the MCP surface exposes, and what it deliberately does not.
For anything this FAQ does not cover, read the docs.
Start pulling clean JSON today.
100 credits to start, no card required. The free tier is the whole API.