evolveguard
About
MCP server wrapping the evolveguard CLI for agent-skill file safety checks.
Details
- Author
- rudrendupaul
- Categories
- Developer Tools
Jump to
Setup
Install evolveguard in your MCP client (Claude Desktop, Cursor, Windsurf, and others).
Repository: https://github.com/rudrendupaul/evolveguard
Follow the installation instructions in the repository README, then restart your MCP client.
What is evolveguard, and why does it exist
evolveguard is a command-line tool and TypeScript library that detects capability drift in Claude Agent Skill files (SKILL.md) and Claude Code auto-memory files (MEMORY.md) after they are edited, by a human or by an agent. It works by parsing a skill's declared frontmatter scope and any static evidence of network or filesystem-write behavior in its body text and bundled hook scripts, snapshotting that as a baseline, and re-deriving the same snapshot after an edit to diff against it. It exists because Claude Code's Agent Skills ecosystem lets skills and memory files change an agent's behavior without a human necessarily reviewing every edit for regression, and no existing tool checks that specific artifact shape without requiring SDK integration or a live agent run.
This is a v0.1 release: a small, focused addition to the existing Claude Agent Skills ecosystem. It ships fully MIT-licensed with no proprietary tier, as two independent, equally first-class packages:
- PyPI (evolveguard-cli, Python), live at](https://github.com/RudrenduPaul/agent-eval)pypi.org/project/evolveguard-cli. A genuine independent port, not a wrapper around the Node binary (seepython/README.md).pip install evolveguard-cliinstalls it directly. The package was originally published under the nameevolveguard; that older PyPI project is retired and no longer receives updates, installevolveguard-cliinstead.
- npm (evolveguard-cli, TypeScript), live atnpmjs.com/package/evolveguard-cli.npm install -g evolveguard-cliinstalls it directly. Renamed 2026-07-19 from the old plainevolveguard, which is now deprecated, to match the PyPI package's naming convention.
What is evolveguard, exactly?A command-line tool and library that detects capability drift in Claude Agent Skill files (SKILL.md) and Claude Code auto-memory files (MEMORY.md) after they are edited. It is not a self-evolving agent framework and does not build, run, or host agents itself. It is a regression-testing CI gate that reacts to a file diff on a skill artifact that already changed, by a human or an agent. See "What is evolveguard, and why does it exist" above for the full definition.
Does evolveguard call an LLM?No. Record and check are both fully static and deterministic; see "Features" above for exactly what each command parses and scans.
What's the core differentiator versus a general testing or eval tool?It needs nothing hosted and nothing to integrate: point it at oneSKILL.mdfile and a fixtures JSON, andrecord/checkwork immediately, with zero SDK integration and no live agent run. That is the tradeoff the "How it compares" table above documents: narrower scope than a general eval platform, in exchange for zero setup.
How does evolveguard compare to Braintrust?Braintrust is a general LLM eval and observability platform that needs SDK integration and an eval-definition step, and it scores real traces from a live agent run. evolveguard needs neither; it parses the skill file itself and never calls an LLM. Use Braintrust if you are already logging traces and want statistical eval scoring across runs. Use evolveguard if you want a pre-commit or CI check that aSKILL.md/MEMORY.mdedit did not silently widen what the skill can do. See the comparison table in "How it compares" above for the full breakdown, including how it compares to this same author'sagent-eval.
Does it work withMEMORY.mdfiles, which have no frontmatter?Yes. A file with no frontmatter is parsed with an empty declared scope, so its capability surface comes entirely from static evidence found in the body text.
What platforms does it run on, and how do I install it?The npm package requires Node.js >=20.12 (any OS Node supports) and installs withnpm install -g evolveguard-cli. The PyPI package requires Python >=3.9 and installs withpip install evolveguard-cli. Both distributions are pure-library/CLI packages with no native bindings, so there is no OS-specific build step on either side.
What's a real limitation to know about before relying on this?It only seesdeclared or showncapability, not runtime behavior. A skill could passcheckand still behave differently on a given prompt in ways that do not touch its capability surface. The false-positive benchmark (see "Features" above) is also currently a small, hand-labeled corpus of 5 before/after pairs, not a large dataset, so treat the 0% figure as a starting measurement, not a statistical guarantee. The Python distribution ships a real MCP server (see "MCP server" above); the npm/TypeScriptmcpsubcommand is still a "coming soon" stub, and the npm build'sevolveguard --versionoutput currently lags the package's real published version (see "CLI command reference" above).
Is this a general agent-evolution framework?No. See "How it compares" above. evolveguard deliberately does not build or host a self-evolving agent framework; it only tests skill/memory edits that already happened.
Is evolveguard free to use, including commercially?Yes. It is MIT-licensed with no proprietary tier or paid version; seeLICENSE. You can use, modify, and redistribute it, including in commercial projects, under the standard MIT terms.
This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.
Create crafted UI components inspired by the best 21st.dev design engineers.
Bring agent evaluations, observability, and synthetic test set generation directly into your IDE for free with Galileo's new MCP server
An MCP server to help AI assistants to answer questions and generate AccelByte Extend SDK code more effectively .
MCP server for AI Diagram Maker — generate beautiful software engineering diagrams directly inside Cursor, Claude Desktop, Claude Code, or any MCP-compatible AI agent
ALAPI MCP Tools,Call hundreds of API interfaces via MCP
AI-powered SVG animation generator that transforms static files into animated SVG components using the Allyson platform
MCP server that gives AI assistants on-demand access to 1,500+ amCharts docs, ~300 code examples, and 1000+ class API references.
APIMatic MCP Server is used to validate OpenAPI specifications using APIMatic. The server processes OpenAPI files and returns validation summaries by leveraging APIMatic’s API.
One shared context layer for AI agents and humans — live API specs, DB schemas, and versioned contracts across repos so every agent and teammate works from the same source of truth.
Build and deploy full-stack Next.js apps with 98 tools for React, AWS, and MongoDB
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
