BroadRouter
Get started

MCP (Model Context Protocol)

Updated

MCP (Model Context Protocol) is an open protocol, built on JSON-RPC 2.0, that standardises how LLM applications connect to external tools and data. An MCP server exposes tools, resources and prompts; an MCP client inside the host application discovers them and passes tool calls from the model to the server.

Also called: Model Context Protocol, MCP server, MCP client, MCP tools

MCP is a plug standard for LLM apps: write a tool server once, and any MCP-aware client can use it.

How it works

The specification describes three roles that talk in JSON-RPC 2.0 messages:

  • Hosts: “LLM applications that initiate connections”, such as an IDE agent or a chat app.
  • Clients: “Connectors within the host application”, one per server.
  • Servers: “Services that provide context and capabilities.”

Servers can offer tools (functions the model may run), resources (data and context) and prompts (message templates). The spec defines two standard transports: stdio for a local subprocess, and Streamable HTTP, where each message is an HTTP POST and replies come back as JSON or a server-sent event stream.

For tools, the client sends tools/list and gets names, descriptions and an inputSchema in JSON Schema. It passes those to the model as tool definitions. When the model returns a tool call, the client sends tools/call with the arguments and hands the result back to the model.

{"jsonrpc": "2.0", "id": 2, "method": "tools/call",
 "params": {"name": "get_weather", "arguments": {"location": "New York"}}}

The model never talks to the MCP server. It only sees tool definitions and results, which is why MCP works with any model that handles tool calls well.

Why it matters

  • Reuse. One GitHub or database server serves every agent and editor that speaks MCP.
  • Token cost. Every listed tool’s name, description and schema is sent as input on each turn. Many servers with many tools make every request longer. Enable only what the task needs.
  • Security. The spec says tools “represent arbitrary code execution” and that hosts must get user consent before invoking any tool. Tool results are also a path for prompt injection.

On BroadNet Inference

BroadNet’s docs do not list an MCP server or MCP-specific endpoint. You run MCP in your client (Cline, Continue, opencode and others) and point the client’s model at BroadNet’s OpenAI-compatible API. BroadNet’s docs list tool calling for GLM 5.3 and Qwen3.8 27B. The MCP with OpenAI-compatible models post shows client setups and a Python bridge.

Questions

Does MCP only work with Claude?

No. MCP is model-agnostic. The client turns MCP tools into ordinary tool definitions, so any model that supports tool calling through your client can use them.

What is the difference between MCP and tool calling?

Tool calling is the model API feature that returns a function name and arguments. MCP is the protocol your app uses to find tools on a server and run them. MCP sits on the application side of the tool loop.

Does BroadNet run an MCP server?

BroadNet's docs do not list an MCP server. You run MCP servers in your own client or agent, and the client sends tool calls to a BroadNet model over the OpenAI-compatible API.

Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity