What is MCP (Model Context Protocol)?
The Model Context Protocol (MCP) is an open standard that connects AI applications to outside tools and data through one common interface.
How it works
Anthropic published MCP in November 2024. An AI application, the client, connects to MCP servers, each of which offers tools the model can call, resources it can read and prompts a person can choose.
Messages are JSON-RPC. A server runs on the same computer over standard input and output, or remotely over Streamable HTTP, where the specification uses OAuth for signing in.
An example
A developer connects Claude Code to a documentation server and a ticket tracker. Asked to fix a bug, Claude Code reads the spec through one server and the ticket through the other, with no code written for either.
Common mistakes
- Connecting a server without reading which tools can write or delete.
- Treating a remote server as safe because the client is trusted.
- Expecting every client to support every part of the specification.
In Nodary
Nodary runs a remote MCP server at https://nodary.ai/api/mcp with 98 tools, 4 prompts and 2 resource templates, on protocol versions 2025-06-18, 2025-03-26 and 2024-11-05. Each tool says whether it only reads and whether a write can overwrite something, so a client can ask you first.
What Nodary does not do: Nodary is an MCP server, not a client: it does not connect to other MCP servers.
Try it on your own documents. The Free plan needs no card.
Create a free workspace