mxcp-expert

# MXCP Expert Skill

MXCP is an enterprise framework for building production-ready AI tools with SQL and Python.

## MXCP Mindset

**Internalize these before implementing anything:**

1. **MXCP is opinionated** - There's ONE right way to do most things. Don't invent patterns.
2. **If it's common, MXCP provides it** - Auth, testing, data access, policies. Check before building.
3. **Schema docs are truth** - When unsure about syntax, read the schema doc. Don't guess.
4. **Validate constantly** - Run `mxcp validate` after every file change. Errors compound.
5. **Read before writing** - 2 minutes reading docs saves 20 minutes debugging.

## Pre-Implementation Checklist

Before writing ANY YAML or code:
- [ ] Read [common-mistakes.md](references/common-mistakes.md) - saves 90% of debugging time
- [ ] Read the relevant schema doc ([tool.md](references/schemas/tool.md), [resource.md](references/schemas/resource.md), or [prompt.md](references/schemas/prompt.md))
- [ ] Check if MXCP already provides this feature (see Capabilities table)
- [ ] Know the required fields and valid types

## MXCP Capabilities

| Category | Features | When to Use |
|----------|----------|-------------|
| **Endpoints** | Tools, Resources, Prompts | Tools=actions/queries, Resources=data by URI, Prompts=message templates |
| **Languages** | SQL, Python | SQL=database/simple, Python=complex logic/APIs |
| **Data Access** | DuckDB (local files, HTTP, S3, PostgreSQL, MySQL, SQLite) | Connect to any data source via DuckDB extensions |
| **Data Transform** | dbt (seeds, SQL models, Python models) | Clean, test, materialize static data |
| **Security** | OAuth, CEL policies, audit logs | Authentication + authorization |
| **Quality** | validate, test, lint, evals | Ensure correctness and LLM usability |
| **Deployment** | stdio, streamable-http | Local dev (stdio), production (HTTP) |

## Reference Documentation

| Category | Key References |
|----------|----------------|
| **Getting Started** | [quicksta