Skip to content

chris-colinsky/forbin-mcp

Repository files navigation

Forbin Logo

CI Python 3.13+ License: MIT Code style: black

Forbin

"This is the voice of world control..." Inspired by Colossus: The Forbin Project, where two computers learn to communicate - just like MCP enables systems to talk to each other.

An interactive CLI tool for testing remote MCP (Model Context Protocol) servers and their tools. Specifically designed for developing agentic workflows with support for suspended services (like Fly.io) that need automatic wake-up.

Companion project: Forbin is the client half of an MCP development workflow. For local dev and eval harnesses, pair it with mock-mcp-server — a config-driven framework for standing up mock MCP servers from OpenAPI specs. See Local development with a mock server.

Name Origin

Forbin is named after Dr. Charles Forbin from the 1970 film Colossus: The Forbin Project. In the movie, two supercomputers (American "Colossus" and Soviet "Guardian") learn to communicate with each other, establishing their own protocol and sharing information - a perfect parallel to the Model Context Protocol enabling AI systems and tools to communicate seamlessly.

Table of Contents

Features

  • Interactive CLI - Browse and test MCP tools with an intuitive command-line interface
  • Automatic Wake-up - Handles suspended services (Fly.io, etc.) with health check probing
  • Cold-Start Resilient - Built-in retry logic and extended timeouts for slow-starting servers
  • Schema Inspection - View detailed tool schemas including parameters and types
  • Generic Tool Calling - Test any MCP tool with interactive parameter input
  • Type-Safe Parameter Parsing - Automatic conversion of strings, booleans, numbers, and JSON objects
  • Connectivity Testing - Verify server connectivity without running tools
  • Clipboard Copy - Press c after any tool response to copy it to the system clipboard

Use Cases

  • Development - Test your FastAPI/FastMCP server tools during development
  • Debugging - Verify tool schemas and responses in real-time
  • Agentic Workflows - Validate tools before integrating them into AI agents
  • CI/CD - Run connectivity tests as part of deployment pipelines
  • Documentation - Explore available tools on any MCP server

Installation

Requirements: Python 3.13 or higher (handled automatically when installing with Homebrew).

End-User Installation

Homebrew (macOS — recommended)

brew tap chris-colinsky/forbin-mcp
brew install forbin-mcp

Prebuilt bottles are published for Apple Silicon (Sequoia, Tahoe). On other platforms, Homebrew falls back to building from source.

pipx (cross-platform)

pipx installs Python applications in isolated environments:

pipx install forbin-mcp

pip

pip install forbin-mcp

After any of the above:

forbin --help

For upgrade, uninstall, and platform-specific notes (including Windows/WSL), see docs/INSTALLATION.md.

Developer Installation

Clone and install in editable mode using uv:

git clone https://github.com/chris-colinsky/forbin-mcp.git
cd forbin-mcp
uv sync

Then run from source:

uv run forbin

For dev dependencies, testing, linting, and pre-commit hooks, see CONTRIBUTING.md and the Development section below.

Configuration

Forbin stores its settings in ~/.forbin/profiles.json — a single document that holds one or more named profiles, each containing one or more named environments. Each environment carries its own MCP_SERVER_URL, MCP_HEALTH_URL, and MCP_TOKEN. Pick a profile/environment per remote server you test against (e.g. howl/prod, howl/preview, internal-api/local).

Option 1 — First-run wizard (easiest): just run forbin. If no profiles file exists, you'll be prompted for the required values and they'll be saved as the default/default profile. Open the in-app editor anytime with forbin --config.

Option 2 — .env seed (one-time): drop a .env in your working directory before the very first launch and Forbin imports the values into the default profile:

cp .env.example .env
MCP_SERVER_URL=https://your-server.fly.dev/mcp
MCP_TOKEN=your-secret-token
MCP_HEALTH_URL=https://your-server.fly.dev/health   # optional — for wake-up

After the seed, the .env is no longer used for connection fields — manage them through forbin --config or p mid-session. Globals (VERBOSE, MCP_TOOL_TIMEOUT) still respect .env and shell vars.

Switching servers — once you have multiple profiles, Forbin shows a picker on launch. Press p mid-session to switch and trigger a reconnect. For scripted runs, pin a target with forbin --profile NAME --env NAME (does not persist). Single-profile users see no picker.

For full details on the schema, migration from v0.1.4, and platform-specific examples, see docs/CONFIGURATION.md.

Configuration Examples

Local Development:

MCP_SERVER_URL=http://localhost:8000/mcp
MCP_TOKEN=test-token-123

Fly.io Production:

MCP_SERVER_URL=https://my-app.fly.dev/mcp
MCP_HEALTH_URL=https://my-app.fly.dev/health
MCP_TOKEN=prod-token-xyz

Usage

Interactive Mode (Default)

Run the interactive tool browser:

forbin

This will:

  1. Show the current configuration and let you confirm or edit it before connecting
  2. Wake up your server (only if MCP_HEALTH_URL is configured — otherwise this step is skipped)
  3. Connect to the MCP server
  4. List all available tools
  5. Enter the two-level interactive browser

Tool List View:

Available Tools

   1. generate_report - Generates a monthly summary report...
   2. get_user_stats - Retrieves user statistics for a given...

Commands:
  [number] - Select a tool
  [v]      - Toggle verbose logging (currently: OFF)
  [c]      - Change configuration
  [p]      - Switch profile / environment
  [q]      - Quit

Tool View:

─────────────────────────── generate_report ───────────────────────────

Commands:
  [d] - View details
  [r] - Run tool
  [v] - Toggle verbose logging (currently: OFF)
  [c] - Change configuration
  [p] - Switch profile / environment
  [b] - Back to tool list
  [q] - Quit

From the tool view you can:

  • d - View full schema with syntax-highlighted JSON
  • r - Run the tool with interactive parameter input
  • b - Go back to the tool list
  • q - Quit

After running a tool, you stay in the tool view to run again with different parameters or navigate elsewhere.

For detailed usage instructions, see the Usage Guide.

Connectivity Test Mode

Test server connectivity without entering interactive mode:

forbin --test

This is useful for:

  • Verifying server is reachable
  • Checking health endpoint configuration
  • Validating authentication tokens
  • CI/CD health checks

Config Editor

Open the in-app config editor at any time to edit the active environment, switch profiles, or manage profile/environment CRUD:

forbin --config

Or pin a target for a single invocation (useful in CI):

forbin --profile staging --env eu-west --test

Help

forbin --help

Local development with a mock server

For local iteration without a real backend, pair Forbin with mock-mcp-server — a config-driven framework that stands up an MCP server from an OpenAPI 3.1 spec with x-mock-* extensions. Useful for agent eval harnesses, deterministic test data, and offline tool browsing.

Two-step workflow:

# 1. In one terminal — start a mock server using a pre-canned config
git clone https://github.com/chris-colinsky/mock-mcp-server.git
cd mock-mcp-server
uv run mock-mcp --config monthly-report

# 2. In another terminal — point Forbin at it
forbin --config   # set MCP_SERVER_URL=http://localhost:8000/mcp, MCP_TOKEN=anything
forbin            # browse and exercise the mock tools

Add a local environment under your project's profile (via the picker's n shortcut) to keep mock-pointing config alongside your prod/preview environments — switch with p mid-session.

How It Works

Forbin is designed to handle the complexities of remote MCP servers, especially those on serverless or suspended platforms.

Health Endpoint Strategy

When MCP_HEALTH_URL is configured, Forbin probes the health endpoint before opening the MCP connection. The probe does two things at once:

  • Availability check — confirms the server is reachable, similar to hitting an LLM provider's /models endpoint to verify the API is up before issuing real requests.
  • Wake-up trigger — on platforms that suspend or stop idle instances (Fly.io scale-to-zero, Railway, Render, etc.), the same request rouses the service.

If you don't configure a health URL, Forbin skips the probe and connects directly — the right choice for always-on servers and local development.

Step Output Colors

During operation, Forbin shows its progress using colored step indicators:

  • [yellow]> Yellow[/yellow]: In Progress - The current action is being performed.
  • [green]+ Green[/green]: Success - The step completed successfully.
  • [dim]- Dim[/dim]: Skip - Step was skipped (e.g., wake-up not needed).

Interactive Toggle

At any time during the connection process or while in the tool menu, you can press v to toggle verbose logging on or off. This is useful for debugging connection issues in real-time without restarting the tool.

Cancelling a Running Tool

While a tool call is in flight, press ESC to cancel it. You stay in the tool view instead of having to ctrl-C the whole CLI — handy for tools that hang or take longer than you're willing to wait.

Terminal Compatibility

Forbin's single-key shortcuts (v, c, p, ESC-to-cancel, post-call clipboard prompt) rely on POSIX termios/tty to read keypresses without requiring Enter. That has a few practical implications:

  • macOS and Linux — fully supported in any modern terminal (Terminal.app, iTerm2, Alacritty, GNOME Terminal, Konsole, etc.).
  • Native Windowstermios isn't available, so the single-key shortcuts silently no-op. Numbered tool selection, prompts, and tool execution still work, but you won't be able to toggle verbose mid-run, cancel a hanging tool with ESC, or use the one-key clipboard prompt. Run Forbin under WSL for the full experience.
  • Piped or non-TTY stdin (e.g. forbin < script.txt, some CI runners) — the shortcuts and the post-call clipboard prompt are skipped automatically. forbin --test is the right mode for non-interactive contexts.
  • Linux clipboard copy — the c shortcut after a tool call requires xclip or xsel to be installed (pyperclip uses them as backends). Without one, Forbin tells you it couldn't access the clipboard and continues.

Detailed Documentation

Development

Project Structure

forbin/
 forbin/              # Package directory
   __init__.py
   __main__.py        # python -m forbin entry point
   cli.py             # Main CLI application
   client.py          # MCP connection + wake-up
   config.py          # Module-level globals, env shadowing, legacy migration
   profiles.py        # profiles.json schema, load/save, profile + env CRUD
   picker.py          # Profile/environment picker UI
   display.py         # Rich-based UI primitives
   tools.py           # Parameter parsing + tool calls
   utils.py           # FilteredStderr + key listeners
   verbose.py         # vlog helpers (gated on VERBOSE flag)
 pyproject.toml       # Python project configuration
 uv.lock              # Dependency lock file
 .env.example         # Example environment configuration
 .env                 # Your environment configuration (not committed)
 CLAUDE.md            # AI assistant guidance
 README.md            # This file

Dependencies

  • fastmcp - MCP client library for Python
  • httpx - Async HTTP client for health checks
  • python-dotenv - Environment variable management
  • pyperclip - Clipboard copy for tool responses
  • rich - Terminal UI rendering

Running Tests

# Test connectivity only
python -m forbin --test

# Run interactive session with your test server
python -m forbin

FastAPI/FastMCP Server Compatibility

This tool is designed to work with FastAPI servers using the FastMCP library. Your server should:

  1. Expose an MCP endpoint (typically /mcp)
  2. Implement bearer token authentication
  3. Optionally expose a /health endpoint for wake-up detection
  4. Follow the MCP protocol specification

Example FastAPI/FastMCP server:

# noinspection PyUnresolvedReferences
from fastapi import FastAPI
# noinspection PyUnresolvedReferences
from fastmcp import FastMCP

app = FastAPI()
mcp = FastMCP("My Tools")

@mcp.tool()
def my_tool(param: str) -> str:
    """A sample tool"""
    return f"Result: {param}"

# Mount MCP at /mcp endpoint
app.include_router(mcp.get_router(), prefix="/mcp")

@app.get("/health")
def health():
    return {"status": "ok"}

Troubleshooting

"Failed to wake up server" or "Failed to list tools: TimeoutError"

  • Verify your MCP_HEALTH_URL is correct
  • Check if the health endpoint is accessible
  • Try removing MCP_HEALTH_URL if your server doesn't suspend
  • For TimeoutError during listing, check if your server is extremely slow or overloaded

"Connection error (server not ready)"

  • Increase the initialization wait time (edit forbin/client.py)
  • Check your MCP_SERVER_URL is correct
  • Verify your MCP_TOKEN is valid

"Session termination failed: 400"

  • This is a harmless error from the FastMCP library
  • Already suppressed in the tool output
  • Safe to ignore

Development

For detailed development instructions, testing, and automation, see docs/DEVELOPMENT.md.

Quick commands:

make install-dev      # Install dev dependencies
make test             # Run tests
make check            # Run all checks (format + lint + test)
make help             # Show all available commands

Testing:

We have comprehensive test coverage with unit and integration tests:

make test             # Run all tests
make test-coverage    # Run with coverage report
make lint             # Check code quality
make format           # Format code

Pre-commit hooks:

Automatically run checks before each commit:

make pre-commit-install

See docs/DEVELOPMENT.md for complete details on testing, CI/CD, and contributing.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for detailed guidelines.

Quick start:

  1. Fork the repository
  2. Create your feature branch (git checkout -b feature/amazing-feature)
  3. Install dev dependencies (make install-dev)
  4. Make your changes and add tests
  5. Run checks (make check)
  6. Commit your changes (git commit -m 'Add amazing feature')
  7. Push to the branch (git push origin feature/amazing-feature)
  8. Open a Pull Request

All pull requests must:

  • Pass all tests (make test)
  • Pass linting (make lint)
  • Maintain or improve code coverage
  • Include appropriate documentation

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

  • Name inspiration: Colossus: The Forbin Project (1970)
  • Built with FastMCP - FastAPI integration for MCP
  • Developed for better MCP tool testing during agentic workflow development

Links

About

Interactive CLI for testing remote MCP (Model Context Protocol) servers. Features automatic wake-up for suspended services, cold-start resilient connections, and an interactive tool browser. Built for developers building agentic workflows.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors