MCP overview
Model Context Protocol (MCP) is an open standard designed to give AI agents a consistent, secure way to discover and use external data, tools, and services. MCP standardizes how applications provide context to LLMs.
MCP overview
There are three components to MCP: the host, client, and server.
- The MCP host is the user-facing application where you interact with the AI. The hosts's job is managing the overall user experience, taking requests, and coordinating the flow of communication.
- The MCP client is the component that handles the actual communication on the host side. You can think of this as a specialized bridge or adapter built into the host application. The host's job is to take the AI's internal request (i.e. "I need to read a file to answer this question") and translate it into a standard MCP request, maintaining a one-to-one connection with an MCP server.
- The MCP server is the gateway to a specific external tool or data source that the AI needs access to. It is a separate program that acts as a secure wrapper around a system like a database, file system, or an external API, like Slack. It's job is to tell the client what it can do ("I can
read_file,query_database, etc."), translate and execute the standardized MCP request from the client, and enforce security by ensuring the AI only accesses what it's allowed to.
There are a few distinctions to make between MCP and APIs.
| Feature | APIs | MCP |
|---|---|---|
| Optimization | Software-to-software communication; deterministic integrations | AI model-to-data communication and agent interactions |
| Implementation | Client (developer) must read documentation and write code to invoke specific endpoints and process the output | Client (agent) can ask the server, "What tools can you offer?" at runtime. The server responds with machine-readable tool descriptions that match the token provided. The same input might result in different output across runs. |
| Output | Machine readable (JSON) and entity IDs | Human readable (markdown) with hydrated names for entities |
MCP and Slack
Slack offers two ways to work with MCP:
- The Slackbot MCP Client connects remote MCP servers to Slack. Once connected, Slackbot discovers your server's tools and invokes them based on user prompts in conversation.
- The Slack MCP Server lets AI apps search channels, send messages, manage canvases, and perform other Slack actions through any MCP-compatible client. Connect it to clients like Cursor and Claude.