Repolode
Submit
highflame-ai

highflame-ai/ramparts

Rust security scanner that audits MCP servers and AI agent skill files for prompt injection, tool poisoning, secret leakage and supply-chain risks.

View on GitHubWebsite
Stars
98
Forks
20
Open issues
16
Last commit
14 days ago

What is it?

Ramparts scans the two surfaces an AI agent trusts most: Model Context Protocol (MCP) servers it connects to, and skill files such as Claude Code slash commands and agentskills.io bundles that it loads and runs by name. It discovers an MCP server's tools, resources and prompts across HTTP, SSE, stdio and subprocess transports, and audits them for prompt injection, tool poisoning, path traversal, command injection and cross-origin escalation.

Skill scanning applies the same pipeline to SKILL.md directories and their bundled scripts. The project combines YARA rules, an LLM-based semantic analyzer, OSV.dev CVE lookups for npx/uvx stdio servers and pinned dependencies, and fingerprint-based drift detection that flags changed MCP configs, tools or skill files after approval. Findings are tagged with the OWASP MCP Top 10 and Agentic Skills Top 10 and can be output as terminal text, JSON, SARIF 2.1.0 or a markdown report.

It is installed with cargo, and can also run as a REST server (`ramparts server`) or as an MCP server over stdio. The project is still under active development and its authors note that it is not perfect and should be combined with runtime guardrails.

Key features

  • 01Scans remote and local MCP servers over HTTP, SSE, stdio and subprocess transports.
  • 02Scans Claude Code commands and agentskills.io bundles, including bundled scripts and references.
  • 03Runs 40 YARA rules with normalized and decoded views to catch obfuscated keywords.
  • 04Checks dependencies against OSV.dev for known CVEs.
  • 05Detects post-approval changes to MCP configs, tools and skill files.
  • 06Exports results as terminal, JSON, SARIF 2.1.0 or markdown reports.

Use cases

  • Vetting third-party MCP servers before connecting an AI assistant to them.
  • Validating agentskills.io skill bundles before publishing them.
  • Gating pull requests that add skills or change MCP configs in CI with SARIF output.
  • Auditing the skills a team has installed across Claude Code, Cursor and similar tools.

Quick start

  1. 1

    Install Ramparts with Cargo

    cargo install ramparts
  2. 2

    Scan a stdio MCP server

    ramparts scan "stdio:npx:mcp-server-commands"
  3. 3

    Auto-discover and scan MCP servers in your IDE configs

    ramparts scan-config

Things to know

  • Ramparts is under active development and is not perfect, so some findings may be inaccurate.
  • It analyzes static metadata, configurations and skill files, so it does not replace runtime guardrails.
  • LLM-based analysis requires additional configuration not covered in the quick start.

Topics

#mcp#ai-agent#security#prompt-injection#static-analysis#rust#sarif#owasp#agent#ai

FAQ

What is ramparts?
Rust security scanner that audits MCP servers and AI agent skill files for prompt injection, tool poisoning, secret leakage and supply-chain risks. Ramparts scans the two surfaces an AI agent trusts most: Model Context Protocol (MCP) servers it connects to, and skill files such as Claude Code slash commands and agentskills.io bundles that it loads and runs by name. It discovers an MCP server's tools, resources and prompts across HTTP, SSE, stdio and subprocess transports, and audits them for prompt injection, tool poisoning, path traversal, command injection and cross-origin escalation.
Is ramparts free to use? What license does it have?
ramparts is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is ramparts still maintained?
Actively maintained. The last commit to ramparts was on Oct 8, 2026.
What can I use ramparts for?
Vetting third-party MCP servers before connecting an AI assistant to them. Validating agentskills.io skill bundles before publishing them. Gating pull requests that add skills or change MCP configs in CI with SARIF output. Auditing the skills a team has installed across Claude Code, Cursor and similar tools.
How do I install ramparts?
Install Ramparts with Cargo: cargo install ramparts Scan a stdio MCP server: ramparts scan "stdio:npx:mcp-server-commands" Auto-discover and scan MCP servers in your IDE configs: ramparts scan-config
Where can I get ramparts?
The source code and installation instructions are on GitHub: https://github.com/highflame-ai/ramparts

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects