MCP server
A server implementing Anthropic's Model Context Protocol, which exposes tools and resources that an LLM can use in its workflows.
Definition
An MCP (Model Context Protocol) server is a service that exposes capabilities to an AI agent via a standardized protocol defined by Anthropic. It can provide tools (functions the agent can call), resources (data the agent can read), and prompts (pre-configured conversation templates). The agent integrates the MCP server into its context and can invoke its tools natively — as if those capabilities were built in. Examples: an MCP server for querying a database, accessing the GitHub API, reading Notion files, or executing SQL queries.
postcursors perspective
MCP is the most interesting mechanism in the current agentic ecosystem because it standardizes extensibility. Before MCP, each tool had its own way of extending agent capabilities (Cursor plugins, VS Code extensions, etc.). With MCP, a compatible server can be used by any compatible client — OpenCode, Kilo Code, Claude Desktop. It's the equivalent of LSP servers for code editors: an interoperability standard.
In practice
Concrete example: you add a PostgreSQL MCP server to OpenCode. The agent can now query your database directly during a debugging session, without you having to copy-paste query results. You tell it "show me the orders with errors since yesterday" and it executes the query itself.
Common misconceptions
- ✗ Thinking MCP servers require complex infrastructure — most are local processes that install with a single npm/pip command
- ✗ Confusing an MCP server with a classic REST API — the protocol is bidirectional and oriented toward LLM interaction