Skip to content

Repository files navigation

toolregistry-server

PyPI version CI License: MIT

Define custom tools and serve them via OpenAPI or MCP interfaces. Built on ToolRegistry.

Overview

toolregistry-server lets you register Python functions as tools and expose them as services through multiple protocols. It provides:

  • Registry Builder: Protocol-agnostic config loading and source registration (registry_builder)
  • Protocol Adapters: OpenAPIAdapter (FastAPI/REST) and MCPAdapter (Model Context Protocol)
  • App Orchestration: App class for building registries and dispatching to any adapter; subclass prepare_registry() for custom registries
  • Authentication: Unified Bearer token support (auth.load_tokens)
  • CLI: toolregistry-server openapi / toolregistry-server mcp with --config, --profile, and more

Ecosystem

Package Description PyPI Docs
toolregistry Core library — tool registration, schema generation, execution PyPI Docs
toolregistry-server Server adapters — expose tools via OpenAPI & MCP PyPI Docs
toolregistry-hub Ready-to-use tools — calculator, web search, file ops, etc. PyPI Docs
toolregistry (core)
       ↓
toolregistry-server (tool server)
       ↓
toolregistry-hub (tool collection + server config)

Installation

# Base (RouteTable, registry_builder, auth)
pip install toolregistry-server

# With OpenAPI support
pip install toolregistry-server[openapi]

# With MCP support
pip install toolregistry-server[mcp]

# Full
pip install toolregistry-server[all]

Quick Start

Programmatic — OpenAPI server

from toolregistry import ToolRegistry
from toolregistry_server import RouteTable
from toolregistry_server.adapters.openapi import OpenAPIAdapter

registry = ToolRegistry()

@registry.register
def greet(name: str) -> str:
    """Greet someone by name."""
    return f"Hello, {name}!"

route_table = RouteTable(registry)
adapter = OpenAPIAdapter(route_table)
adapter.run(host="0.0.0.0", port=8000)

Programmatic — MCP server

import asyncio
from toolregistry import ToolRegistry
from toolregistry_server import RouteTable
from toolregistry_server.adapters.mcp import MCPAdapter

registry = ToolRegistry()
# ... register tools ...
route_table = RouteTable(registry)

adapter = MCPAdapter(route_table)
adapter.run(transport="stdio")                        # blocking
# or: asyncio.run(adapter.run_async(transport="sse", host="0.0.0.0", port=8000))

High-level — App class

from toolregistry_server.app import App

# From a config file
App().serve_openapi(config_path="tools.yaml", host="0.0.0.0", port=8000)
App().serve_mcp(config_path="tools.yaml", transport="stdio")

# From a pre-built registry
from toolregistry import ToolRegistry
registry = ToolRegistry()
# ... register tools ...
App().serve_openapi(registry=registry, port=9000)

Custom App subclass

Override prepare_registry to add built-in tools, hooks, or metadata:

from toolregistry_server.app import App

class MyApp(App):
    def prepare_registry(self, **kwargs):
        from toolregistry import ToolRegistry
        registry = ToolRegistry()
        registry.register(my_builtin_tool)
        # optionally apply user config on top
        if kwargs.get("config_path"):
            from toolregistry_server import apply_config, load_config
            apply_config(registry, load_config(kwargs["config_path"]))
        return registry

MyApp().serve_openapi(host="0.0.0.0", port=8000)

CLI

# OpenAPI server from config file
toolregistry-server openapi --config tools.yaml --port 8000

# MCP server (stdio)
toolregistry-server mcp --config tools.yaml --transport stdio

# MCP server (SSE)
toolregistry-server mcp --config tools.yaml --transport sse --port 8000

# With deployment profile (disables network/filesystem tools)
toolregistry-server openapi --config tools.yaml --profile remote

# With Bearer token auth
toolregistry-server openapi --config tools.yaml --tokens /path/to/tokens.txt

Config File

JSONC and YAML are both supported. Three source types: python, mcp, openapi.

mode: denylist     # or "allowlist"
disabled: []       # namespaces to exclude (denylist mode)

tools:
  # Python module — all public functions
  - type: python
    module: my_package.tools
    namespace: my_tools

  # Python class
  - type: python
    class: my_package.Calculator
    namespace: calculator

  # MCP server (stdio subprocess)
  - type: mcp
    transport: stdio
    command: ["python", "-m", "my_mcp_server"]
    namespace: mcp_tools

  # MCP server (SSE / streamable-http)
  - type: mcp
    transport: http
    url: http://localhost:8080/mcp
    namespace: remote_mcp

  # OpenAPI endpoint
  - type: openapi
    url: https://api.example.com/openapi.json
    namespace: external_api
    auth:
      type: bearer
      token_env: EXTERNAL_API_TOKEN

See examples/config.yaml and examples/config.jsonc for full examples.

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    registry_builder                         │
│   load_config · apply_config · register_*_source            │
│   apply_profile · PROFILE_DISABLE_TAGS                      │
└─────────────────────────┬───────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────────┐
│                       RouteTable                            │
│              (central routing layer)                        │
└─────────────────────────┬───────────────────────────────────┘
                          │
          ┌───────────────┼───────────────┐
          ▼               ▼               ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│  OpenAPIAdapter │ │   MCPAdapter    │ │  (your adapter) │
│   (FastAPI)     │ │  stdio/sse/http │ │  Adapter ABC    │
└─────────────────┘ └─────────────────┘ └─────────────────┘
          │               │
          ▼               ▼
┌─────────────────┐ ┌─────────────────┐
│  HTTP Clients   │ │   MCP Clients   │
└─────────────────┘ └─────────────────┘

Adding a New Adapter

  1. Subclass Adapter from toolregistry_server.adapters
  2. Implement run(**kwargs) and create_and_run(cls, route_table, **kwargs)
  3. Optionally implement add_cli_arguments(parser) for CLI integration
  4. Call App().serve(MyAdapter, ...) — no changes to App needed

Deployment Profiles

--profile applies tag-based tool filtering at startup:

Profile Disables
remote FILE_SYSTEM, DESTRUCTIVE, PRIVILEGED tagged tools
local NETWORK tagged tools

Documentation

Contributing

Contributions are welcome! Please see our Contributing Guide for details.

License

MIT — see LICENSE.

Related Projects

About

ToolRegistry Server: Serve Custom Tools via OpenAPI and MCP for Function-Calling LLMs

Topics

Resources

Contributing

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages