packmcp

PackMCP helps you build smaller, safer MCP toolsets for real tasks.

Instead of handing every tool to the agent, PackMCP lets you:

• inspect a tools/list manifest or raw tool array
• estimate how much tool metadata costs in context tokens
• spot risky write, merge, deploy, or dispatch tools early
• score tools against a concrete task and pack profile
• generate a reusable PackMCP pack JSON with only the selected tools
• replay a saved pack or allowlist against a newer manifest and detect drift
• fingerprint saved tool definitions so drift checks catch schema changes, not just missing names
• export a tighter allowlist for Python and TypeScript runtimes
• run the same analysis from the CLI for scripts and CI
• compare two MCP manifests before migration or rollout
• analyze a real mcp.json server through the official MCP Inspector CLI

As of 2026-03-20, MCP is already broad enough that the problem is no longer just connection setup. The more common problem is tool overload:

• too many tools in the prompt
• too many side-effecting actions exposed by default
• too little clarity about what the agent actually needs

PackMCP is built for that layer.

• local-first static UI
• task-aware tool scoring
• risk classification and warning cards
• recommended pack generation by profile and risk budget
• copyable exports for allowlists and SDK filters
• reusable pack artifact export for downstream enforcement or review flows
• saved pack replay with drift warnings and optional strict CI failure mode
• definition-level drift detection for saved packs and Inspector-backed reviews
• optional multi-manifest comparison mode
• official MCP Inspector CLI integration for live server analysis
• sample manifest and testable core logic

npm start

Then open `http://