Skip to content

Releases: gabrielekarra/mcp-anything

v0.2.0 — Domain pipeline

Choose a tag to compare

@gabrielekarra gabrielekarra released this 29 Apr 09:59
d0f17b2

What's new

mcp-anything build --brief — brief-driven generation

Write a short YAML file describing your domain, use cases, and data source. The LLM pipeline does the rest:

mcp-anything build --brief my-api.yaml -o ./my-mcp-server

Five LLM phases run end-to-end: domain modeling → tool design → emit → skill bundle → validation harness.

2026 tool design rules

  • Group CRUD: ≥3 operations on the same resource → single manage_X(operation=...) tool
  • Full lifecycle: create, read, update, delete, list, confirm, capture, cancel, finalize, void — all included
  • Compact + verbose: every tool has a verbose flag
  • Discovery endpoint: every server exposes GET /.well-known/mcp

Validated against real APIs

  • Stripe: 13 grouped tools — covers 10+ capabilities absent from the official 15-tool Stripe agent toolkit
  • GitHub: 22 grouped tools — 100% in-scope coverage, matching the official 51-tool server at 57% fewer tools

Installation

pip install -U mcp-anything

Quick start

export ANTHROPIC_API_KEY=sk-...
mcp-anything build --brief my-api.yaml -o ./my-mcp-server
cd my-mcp-server && pip install -e . && python -m mcp_<name>.server

See the README for full documentation.

v0.1.0 — First public release

Choose a tag to compare

@gabrielekarra gabrielekarra released this 08 Apr 19:47

v0.1.0 — First public release

MCP-Anything auto-generates fully functional MCP server packages from any codebase. Point it at source code — Python, Java, Go, Rust, Ruby, TypeScript — and get a pip-installable MCP server in minutes, no manual wrapping required.

Key capabilities

  • 15 static detectors — CLI (argparse, Click, Typer), HTTP frameworks (FastAPI, Flask, Spring Boot, Express, Django, Rails, Gin, Actix, Axum), OpenAPI/Swagger, GraphQL, gRPC/Protobuf, WebSockets
  • 6-phase pipeline — Analyze, Design, Implement, Test, Document, Package — with JSON manifest for resume support
  • Multiple backend strategies — CLI subprocess, HTTP proxy, Python call, stub — auto-selected based on what the codebase exposes
  • Full MCP protocol — Tools, Resources, Prompts, AGENTS.md generation for agent discoverability
  • HTTP transport--transport http for remote deployment with OpenTelemetry tracing and Docker packaging
  • LLM-enhanced analysis — Optional Claude API integration for better tool descriptions and grouping

Installation

pip install mcp-anything

Quick start

mcp-anything generate /path/to/any/app

See the README for full documentation, examples, and usage.

What's next

  • Desktop software showcase (Blender, GIMP, Audacity)
  • OAuth/token auth on the MCP server itself
  • Plugin system for custom detectors