bellwether
Open-source testing tool for Model Context Protocol (MCP) servers. Detect breaking schema changes before they reach production. Free deterministic validation with optional LLM-powered behavioral exploration. CI/CD ready with GitHub Actions support.
claude mcp add --transport stdio dotsetlabs-bellwether npx -y @dotsetlabs/bellwether
How to use
Bellwether is the open-source MCP testing tool designed to catch drift in tool schemas used by AI assistants via the Model Context Protocol. It helps you automatically explore and test MCP tool interfaces, compare against baselines, and surface changes that could cause downstream failures in production. The core workflow focuses on drift detection in CI or local development; advanced workflows enable deeper behavioral testing and contract verification. You can initialize a server, run checks to detect schema drift, and optionally save baselines for future comparisons. Bellwether supports both deterministic drift checks and optional exploratory testing to surface edge cases across MCP tools and prompts.
How to install
Prerequisites:
- Node.js and npm installed on your system
- Git access to clone or install via npm registry
Step-by-step installation:
- Install Bellwether globally (recommended for quick start): """ npm install -g @dotsetlabs/bellwether """
- Verify installation: """ bellwether --version """
- Initialize a server for drift testing (example using MCP server placeholder): """ bellwether init npx @mcp/your-server """
- Run a drift check and optionally save a baseline: """ bellwether check bellwether baseline save """
- Integrate into CI by adding Bellwether commands to your workflow as shown in the documentation examples.
Additional notes
Notes and tips:
- Bellwether is used to validate MCP tool schemas and detect drift between current tool behavior and a saved baseline.
- You can run in different modes: check (drift detection) and explore (LLM-driven behavioral testing), depending on your needs.
- Environment variables for external AI services (OpenAI, Anthropic, Ollama) are only needed if you enable exploration features; they are not required for core drift checks.
- When using remote MCP servers, you can override headers or pass authentication through the bellwether.yaml configuration or CLI overrides as documented.
- Ensure baselines are committed to your CI to enable consistent drift comparisons across PRs.
Related MCP Servers
awesome-agent-skills
A curated list of skills, tools, tutorials, and capabilities for AI coding agents (Claude, Codex, Antigravity, Copilot, VS Code)
claude-code-plugins-plus-skills
270+ Claude Code plugins with 739 agent skills. Production orchestration patterns, interactive tutorials (11 Jupyter notebooks), and CCPI package manager. 4.13.0
claude-emporium
🏛 [UNDER CONSTRUCTION] A (roman) claude plugin marketplace
automagik-genie
🧞 Automagik Genie – bootstrap, update, and roll back AI agent workspaces with a single CLI + MCP toolkit.
forge-orchestrator
Forge Orchestrator: Multi-AI task coordination. File locking, knowledge capture, drift detection. Rust.
mcp-task-orchestrator
A Model Context Protocol server that provides task orchestration capabilities for AI assistants