Skip to main content

Turn any Python codebase into a production-grade MCP server with zero friction.

Project description

image

Smarter-MCP 🚀

The highest-level Python framework for building, generating, and running MCP servers.

If Python can call it, Smarter-MCP can serve it.

pip install smarter-mcp

Smarter-MCP sits on top of FastMCP and acts as the orchestration layer that handles everything a developer shouldn't have to think about:

Three ways in, one runtime

1. Existing code — zero rewrites

Point at any module you already have (or anything on PyPI):

import pandas as pd
from smarter_mcp import SmarterMCP

app = SmarterMCP("data-tools")
app.discover_module(pd.DataFrame, include=["describe", "head", "tail"])
app.run()

Or scan an entire local codebase from the CLI:

smarter-mcp serve ./src/mylib

The dual-pass engine (AST + inspect) reads your signatures, builds JSON schemas, and serves them. Nothing to rewrite.


2. Stateful class tools

When your tools need shared state (a DB connection, an API client, an ML model loaded once), @toolkit manages it:

from smarter_mcp import tool, toolkit

@toolkit(lifecycle="session")
class DatabaseClient:
    def __init__(self, host: str = "localhost", port: int = 5432):
        self.conn = connect(host, port)

    @tool(name="run_query")
    def query(self, sql: str) -> list[dict]:
        """Execute a SQL query and return results."""
        return self.conn.execute(sql).fetchall()

One instance per session. Constructor args injected from config. Cleaned up on session end. You write the class, Smarter-MCP handles the plumbing.


3. ✍️ Fresh tools from scratch

from smarter_mcp import SmarterMCP, tool, resource

app = SmarterMCP("my-server")

@tool("Greet a user by name")
def greet(name: str) -> str:
    return f"Hello, {name}!"

@resource("config://settings")
def get_settings() -> dict:
    return {"debug": True, "version": "1.0"}

app.run()

⚙️ What the runtime gives every tool

Regardless of how a tool was registered, every call goes through:

  • Schema validation — parameters validated before your function is called. Clean errors back to the agent, not raw tracebacks. Type coercion — agents send "42" instead of 42 constantly. Handled silently.
  • Schema validation — parameters validated before your function is called. Clean errors back to the agent, not raw tracebacks.
  • Schema validation — parameters validated before your function is called. Clean errors back to the agent, not raw tracebacks.
  • MultimodalPIL.Image and np.ndarray parameters decoded from ImageContent automatically. Return images and they're wrapped back up.
  • Instance lifecyclesession, singleton, or per-call. Resolved and bound per request.
  • Namespace routing — auto-derived from module paths. No silent name collisions.
  • Auth + rate limiting — API key middleware and sliding-window rate limits, config-driven.

✨ AI-generated tool descriptions

Most Python code in the wild has no docstrings. That's fine.

Point Smarter-MCP at an undocumented library and it will write the tool descriptions for you using Claude, GPT, or any OpenAI-compatible model. Every tool your agent sees gets a clean, accurate description regardless of what the original code looked like.

llm:
  provider: anthropic        # or openai, openrouter
  model: claude-3-5-haiku
  cache: true

or in code:

server = SmarterMCP(
    "my-server",
    llm_enabled=True,
    llm_provider="anthropic",
    llm_model="claude-3-5-haiku",
)
server.run()

Undocumented code stops being a blocker.


📋 YAML manifest — no Python required

name: my-server
version: 0.1.0

server:
  host: 0.0.0.0
  port: 8000
  transport: sse

sources:
  - path: ./src/my_local_utils
  - module: random
    include: [choices, randint]
    namespace: random_tools

expose:
  private: false
  unannotated: warn
smarter-mcp serve --manifest smarter-mcp.yaml

🚀 Getting started

pip install smarter-mcp

For multimodal support (PIL / numpy):

pip install "smarter-mcp[multimodal]"

Then point it at your code:

smarter-mcp serve ./src/mylib --port 8000
smarter-mcp validate ./my_tools.py
smarter-mcp test ./my_tools.py

Built on FastMCP.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

smarter_mcp-0.1.1.tar.gz (214.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

smarter_mcp-0.1.1-py3-none-any.whl (73.0 kB view details)

Uploaded Python 3

File details

Details for the file smarter_mcp-0.1.1.tar.gz.

File metadata

  • Download URL: smarter_mcp-0.1.1.tar.gz
  • Upload date:
  • Size: 214.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.11.19 {"installer":{"name":"uv","version":"0.11.19","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for smarter_mcp-0.1.1.tar.gz
Algorithm Hash digest
SHA256 3e2ef73b5e9f81f7c0d5423702d7300ac628d9e222d92d95018e6cfea6f72d86
MD5 05e7f32ed9418b9a7524ce33ebfdc7fb
BLAKE2b-256 31a08d1f61950929a81f4839f400d44cefb844fa4c69380378b9554201d7dadd

See more details on using hashes here.

File details

Details for the file smarter_mcp-0.1.1-py3-none-any.whl.

File metadata

  • Download URL: smarter_mcp-0.1.1-py3-none-any.whl
  • Upload date:
  • Size: 73.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.11.19 {"installer":{"name":"uv","version":"0.11.19","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for smarter_mcp-0.1.1-py3-none-any.whl
Algorithm Hash digest
SHA256 8f1aea0853182850d802616b8da5935b536553904d12dd6e9ff03ad39bcb62b7
MD5 3bfe598fdfad0fcfc123c3e65a3dcb83
BLAKE2b-256 94b36bfdbaf57e19311c76a7b17b304c238e8f09eaaf5e482a1099328390506c

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page