Pharo MCP Server

by paulwilke

14 stars
241 downloads
Not rated
GitHub

About

Pharo Smalltalk implementation of Anthropic's Model Context Protocol (MCP) server specification. Enables LLMs like Claude to interact with Pharo applications.

Details

Author
paulwilke
GitHub stars
14
Downloads
241
Categories
Other

- Implements core MCP endpoints: discovery, service description, and invocation.
- Dynamic registration of Pharo methods or blocks as tools.
- Automatic generation of service‑description JSON from registered tools.
- Built on the Teapot micro web framework and NeoJSON for JSON handling.
- Extensible for integration into larger Pharo applications.
- Planned input validation using JSON Schema and helpers for common object models.

Install the server using Metacello with the provided code, then register Pharo methods or blocks as tools using the built‑in API. The server exposes configurable HTTP endpoints for discovery, description, and invocation.

Pharo MCP Server

License: MIT
Certified by MCP Review

<!-- Add other badges later, e.g., build status, version -->

Pharo Smalltalk implementation of Anthropic's Model Context Protocol (MCP) server specification. Enables LLMs like Claude to interact with Pharo applications.

---

Table of Contents

What is MCP?
Goal of this Project
Features
Installation
Basic Usage
Defining Tools
Configuration
Development & Contributing
License

What is MCP?

The Model Context Protocol (MCP), developed by Anthropic, is a specification designed to standardize how Large Language Models (LLMs) interact with external tools and services. It defines mechanisms for:

1. Discovery: How an LLM finds out if a service supports MCP and gets basic information.
2. Service Description: How a service describes its available tools (functions), including their names, purposes, input parameters, and output formats.
3. Invocation: How an LLM securely calls a specific tool with the required arguments and receives the result.

The goal is to make LLM tool usage more reliable, secure, and standardized.

For more details, see the official Model Context Protocol GitHub organization.

Goal of this Project

This project aims to provide a robust and easy-to-use server-side implementation of the MCP specification specifically for the Pharo Smalltalk environment.

It allows developers to expose functionalities within their Pharo applications (like querying data, creating objects, triggering actions) as "tools" that MCP-compliant LLMs (such as Claude 3) can discover and invoke through the standardized protocol.

This bridges the gap between the dynamic Pharo environment and the capabilities of modern LLMs.

Features

MCP Specification Compliance: Implements the core server-side endpoints:
Discovery (/.well-known/model-context)
Service Description (/mcp/description - or configurable path)
Invocation (/mcp/invoke - or configurable path)
Dynamic Tool Registration: Provides a simple API to register Pharo methods or blocks as MCP tools.
Automatic Service Description: Generates the MCP Service Description JSON based on the registered tools and their metadata (name, description, schemas).
Web Server Integration: Built upon the Teapot micro web framework.
JSON Handling: Uses NeoJSON for robust JSON parsing and generation.
Extensible: Designed to be integrated into larger Pharo applications.
(Planned): Input validation using JSON Schema definitions for tool parameters.
(Planned): Integration helpers for common Pharo object models (e.g., using Voyage).

Installation

This project is intended to be loaded using Metacello.

1. Ensure Metacello is loaded in your Pharo image (it usually is).
2. Execute the following code in a Pharo Workspace (adjust the repository URL and branch/tag if necessary):

```smalltalk
Metacello new
baseline: 'MCPFramework'; "<- Use the actual name of YOUR baseline class (without 'BaselineOf')"
repository: 'github://YOUR_USERNAME/pharo-mcp-server:main/'; "<- CHANGE THIS to your repo URL and branch/tag"
load.

No reviews yet — be the first

Sign in to leave a review

Use Google, GitHub, or an email account so ratings stay tied to real people.

Email sign in

No reviews posted yet.