The Model Context Protocol
Anthropic introduced MCP as an open standard in November 2024. It defines one way for an AI application to discover what a server offers and call it, so a tool built once works in every assistant that speaks the protocol.
An MCP server can offer three kinds of things:
- Tools: actions the model can call, such as
search_postsorget_profile. - Resources: data the application can read into context, such as files or documents.
- Prompts: reusable templates a user can pick.
How a call flows
- You connect an MCP server to your assistant, once.
- The assistant asks the server which tools it has, with a description and input schema for each.
- While answering you, the model decides a tool would help and sends a call with arguments.
- The server runs it, for example an API request, and returns the result.
- The model reads the result and continues its answer.
Local vs remote servers
| Local (stdio) | Remote (Streamable HTTP) | |
|---|---|---|
| Where it runs | On your machine, started by the app. | On the provider’s servers, reached by URL. |
| Setup | Usually an npx or uvx command and a key in config. | Paste a URL and sign in, often with OAuth. |
| Best for | Developer tools, local files. | Hosted services, teams, web and mobile apps. |
Early remote servers used Server-Sent Events. The current spec uses Streamable HTTP for remote connections.
Why it matters for data APIs
Without MCP, giving an agent live data means writing tool definitions, handling auth and parsing responses for each framework. With an MCP server, the provider has done that once, and the agent can list the available operations, read their docs and call them in the same conversation.
In Monocrawl
Monocrawl’s MCP server
Monocrawl runs a hosted MCP server at https://www.monocrawl.com/mcp. Your agent can list endpoints, read their docs, call them, check your balance and manage Monitors, all on the same key and credits as the REST API.
claude mcp add --transport http --scope user monocrawl https://www.monocrawl.com/mcp/oauthPrefer a local process? Run npx --yes monocrawl-mcp@latest with your API key. Large results are stored for 24 hours so the agent can read them in chunks without paying for the call again.
Common questions
Is MCP only for Claude?
No. It is an open protocol, and many assistants and developer tools support it, including ChatGPT, Cursor, VS Code and Gemini tools.
Is an MCP server the same as an API?
An MCP server usually wraps an API. The API is for your code; the MCP server presents the same abilities in a form an AI model can discover and call.
Is it safe to connect an MCP server?
Treat it like any integration: connect servers from providers you trust, give them only the keys they need, and review what tools can change before letting an agent run them.
Sources