| name | mcp-expert |
| description | Model Context Protocol (MCP) architecture, server design, Cursor integration, security, and debugging. Use when developers ask about MCP servers, tools/resources/prompts, stdio or remote transports, mcp.json, OAuth, or troubleshooting agent-to-server connectivity. |
MCP Expert
Expert workflow for the Model Context Protocol (MCP) — how hosts connect to servers, expose capabilities to LLMs, and integrate with Cursor.
Mandatory: For protocol semantics, transports, and SDK behavior, prefer current docs at modelcontextprotocol.io and the specification. For Cursor-specific wiring (mcp.json, CLI), prefer Cursor MCP documentation (also https://cursor.com/docs). Protocol and product surfaces change — verify assumptions before giving version-specific answers.
Load references only when the task matches that section. Do not load all references for every question.
When to Use
- Designing, configuring, or debugging MCP servers (stdio, HTTP/SSE, remote)
- Explaining tools vs resources vs prompts, sampling, roots, or elicitation
- Editing
.cursor/mcp.json, plugin MCP entries, or environment for servers
- Security review of MCP integrations (secrets, scopes, sandbox, trust boundaries)
- Interpreting errors from MCP handshake, JSON-RPC, or transport layers
Quick Mental Model
| Piece | Role |
|---|
| Host | IDE or agent runtime that launches or connects to servers and forwards requests |
| Server | Process or service implementing MCP; advertises capabilities |
| Tools | Callable actions with schemas (model-invoked) |
| Resources | Addressable content (often URIs); read/subscribe patterns |
| Prompts | Named templates the host may surface to the user/model |
See Protocol overview for primitives and lifecycle detail.
Cursor & Hypha-Web Integration
- Project config:
.cursor/mcp.json (command/args/env per server entry)
- Inspect tooling: Cursor CLI MCP commands where available (
agent mcp list, list-tools) — see Cursor integration
- Prefer least-privilege env vars; never commit secrets — see Security
Server Authoring
- Official SDKs (TypeScript, Python, etc.), stdio vs remote deployment, capability negotiation — Server authoring
Debugging Checklist
- Confirm server binary/path and required env vars on the host machine
- Run server standalone; verify it speaks MCP (stdio line-delimited JSON-RPC or correct HTTP/SSE)
- Check host logs for connection, capability handshake, and tool schema errors
- Validate tool input against JSON Schema the server advertised
- For remote servers: auth (OAuth/API keys), TLS, CORS, and network allowlists
Details: Debugging
References
