The MCP server for every kind of data.
An MCP server gives AI agents tools they can call. Connect PrimaryAPI’s to Claude Code, Cursor, Codex or VS Code, and your agent can search the catalog, read an API’s schema and price, and run it with your key.
Server URL
claude mcp add --transport http primaryapi \
https://api.primaryapi.com/mcp \
--header "Authorization: Bearer $PRIMARYAPI_API_KEY"Run it in a terminal. Your shell fills in the key from the environment.
What it is
What is an MCP server? A set of tools an agent can call.
MCP, the Model Context Protocol, is an open-source standard for connecting AI applications to external systems. Its docs compare it to a USB-C port for AI applications. Read on What is the Model Context Protocol? (modelcontextprotocol.io) (opens in a new tab)
Through MCP, AI applications such as Claude or ChatGPT reach data, tools such as search engines and calculators, and workflows such as specialised prompts. Read on What is the Model Context Protocol? (modelcontextprotocol.io) (opens in a new tab)
An MCP server is the program that offers those tools. Each tool has a name, a description and the arguments it takes, and the AI client decides when to call one and with what.
MCP defines two standard transports: stdio, where the client launches the server as a local subprocess, and Streamable HTTP, where each message is an HTTP POST to one endpoint. Read on MCP specification: transports (modelcontextprotocol.io) (opens in a new tab)
PrimaryAPI’s server is a remote one over Streamable HTTP: your client sends its requests to one URL, with your key in a header.
The tools
What your agent can do with PrimaryAPI’s server.
The same verbs as the REST API, the same prices and the same errors. Arguments are strict: an API’s own fields go inside input.
| Tool | What it does | Key |
|---|---|---|
| search_apis | Find APIs by keyword or use case. | No key |
| list_apis | Browse APIs by category or platform, a page at a time. | No key |
| get_api | Read one API’s input and output schema, examples, notes and price. | No key |
| quote_api | Check an input and see its price in credits without running it. | Run or full |
| run_api | Run an API and get its result. This spends credits. | Run or full |
| get_request | Check on an earlier call, with its result once it is done. | Any key |
| read_result | Read a stored result again for free, trimmed if you like. | Any key |
| get_balance | See the workspace’s credits and whether its keys are paused. | View or full |
| report_bug | Report wrong data or an error that looks like ours. | Any key |
| send_feedback | Send feedback, or ask for an API that is missing. | Any key |
How an agent uses it
Find it, read it, then run it.
search_apisSearch for what it needs, such as “google maps reviews”. Free.get_apiRead the chosen API’s input schema before its first run: field names are exact.quote_apiOptionally, check the price for this exact input. Free.run_apiRun it, asking only for the fields and items it needs.get_requestIf the call is still running, check back later instead of running it again, which would start a second job.
// 1. Find an API. No key needed.
{
"name": "search_apis",
"arguments": {
"query": "reddit posts about a topic"
}
}
// 2. Run it with your key.
{
"name": "run_api",
"arguments": {
"sku": "reddit.search",
"input": {
"query": "best mechanical keyboard",
"limit": 10
},
"fields": [
"title",
"score",
"permalink"
],
"max_items": 5
}
}By platform
Platforms your agent can read.
One server and one key reach every platform in the catalog. These pages show what an agent gets for each one.
Spending and safety
Agents spend only what you allow.
Browse without a key. An agent that only looks things up can connect with no key at all.
Cap what it can spend. Give each agent its own key with run access and a spend limit, so a runaway loop stops at a number you chose.
Keep the key out of prompts. Set it in the client’s configuration. The server never asks a model for its key and never returns one.
Treat results as data. Posts, comments and pages are written by other people. Don’t let what an API returns give your agent instructions.
Restricted APIs stay closed. APIs that return personal contact details need an approved account, and an agent’s trial key can never call them.
What is an MCP server?
A program that offers tools to AI applications over the Model Context Protocol, an open standard. An agent connected to one can see its tools, read what each one needs and call them while it works.
What does PrimaryAPI’s MCP server do?
It puts the PrimaryAPI catalog in front of your agent. The agent can search for an API, read its schema and price, run it with your key and re-read results, across social platforms, maps, company filings and validation checks.
Which MCP clients does it work with?
Claude Code, Claude Desktop, Cursor, Codex and VS Code, with the set-up shown on this page, and any other client that supports Streamable HTTP with custom headers.
Is the MCP server free?
Connecting is free, and the tools that search and describe APIs work without a key. Running an API spends credits at that API’s price, and you only pay when data comes back.
Do I need an API key for the MCP server?
Only to run APIs, check prices or read your balance. Give agents a key with run access and a spend limit, set in the client’s configuration and never in a prompt.
What is the difference between a local and a remote MCP server?
A local server runs on your computer: the client starts it and talks to it over standard input and output. A remote server runs elsewhere and the client reaches it over HTTP. PrimaryAPI’s server is remote.
Can an AI agent sign itself up?
Yes. An agent can mint a capped trial key on its own, use it right away and hand you a claim link when it needs more. You approve a spend limit, and the agent keeps the same key.
Hand your agent every kind of data.
Create an account, make a key with a spend limit, and add one URL to your MCP client.