convez.ai
MCP ServerSegurançaIntermediário
RP-Digital-Innovations

Context Snipe

RP-Digital-Innovations/context-snipe

Relatório de CVEs determinístico para assistentes de IA via MCP

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP em Rust que lê lockfiles do seu projeto, valida cada dependência contra OSV.dev e fornece um relatório de vulnerabilidades preciso e priorizado para seu assistente de IA. Elimina hallucinations substituindo suposições por análise determinística de suas dependências reais.

Instalação

$npx convez-ai RP-Digital-Innovations/context-snipe

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
v0.3.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Análise de vulnerabilidades em dependências reais do projeto
  • Integração de segurança em workflows de coding assistants (Claude, Cursor, Windsurf)
  • Redução de ruído: apenas CVEs que afetam pacotes efetivamente instalados
  • Contextualização precisa de IA sobre risco de segurança do codebase

Veredito

Ferramenta focada e bem executada que resolve um problema real: assistentes de IA sem visibilidade determinística das dependências do projeto. Leve (~0.85 MB), rápida em Rust, e integração direta com OSV.dev torna-a prática para pipelines de segurança em tempo de desenvolvimento.

README

A tradução para o português deste README ainda está na fila.

CI License: MIT GitHub release MCP

Works with: Claude Desktop · Cursor · Windsurf · VS Code · Zed · any MCP client


$ context-snipe scan .

context-snipe — vulnerability scan
Project: ./my-api
Scanned: 412 entries (388 unique packages) from Cargo.lock, package-lock.json

FOUND  3 advisories affecting 2 of 388 package(s):

  lodash 4.17.11  [npm]
    [CRIT] CVE-2019-10744  Prototype Pollution in lodash
    [HIGH] CVE-2021-23337  Command Injection in lodash

  minimatch 3.0.4  [npm]
    [HIGH] CVE-2022-3517   minimatch ReDoS vulnerability

Source: OSV.dev — packages actually in your resolved dependency tree.

The problem

You ask Cursor or Claude: "Does my project have any security issues?"

It doesn't know your packages. It doesn't know your versions. It hallucinates an answer based on general knowledge — not your actual package-lock.json.

Your scanner (Dependabot, Snyk, whatever) floods you with 200 warnings, most of which don't apply to what you actually ship. You spend 45 minutes Googling CVEs that are irrelevant to your code.

context-snipe closes both gaps. It reads your resolved lockfiles (not your package.json — your actual installed packages), asks OSV.dev only about what you have, deduplicates the noise, ranks by real CVSS severity, and gives your AI a clean, accurate briefing it can actually reason about.


Install in 30 seconds

macOS / Linux — one line, picks the right binary for your platform:

curl -fsSL https://raw.githubusercontent.com/RP-Digital-Innovations/context-snipe/main/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/RP-Digital-Innovations/context-snipe/main/install.ps1 | iex

Rust users — from crates.io:

cargo install context-snipe        # build from source
cargo binstall context-snipe       # or grab the prebuilt binary, no compile

Grab the binary for your platform from the latest release:

PlatformAsset
macOS (Apple Silicon)context-snipe-aarch64-apple-darwin
macOS (Intel)context-snipe-x86_64-apple-darwin
Linux x86_64context-snipe-x86_64-linux
Linux ARM64context-snipe-aarch64-linux
Windows x86_64context-snipe-x86_64-pc-windows.exe

chmod +x it and move it onto your PATH.

Verify:

context-snipe --version

Add to your AI tool (60 seconds)

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "context-snipe": {
      "command": "context-snipe",
      "args": ["serve"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "context-snipe": {
      "command": "context-snipe",
      "args": ["serve"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "context-snipe": {
      "command": "context-snipe",
      "args": ["serve"]
    }
  }
}

Restart your editor. Then ask your AI: "Check this project for vulnerable dependencies."


What your AI can now do

MCP ToolWhat it does
scan_dependenciesLists every resolved package in your project (name, version, ecosystem)
check_vulnerabilitiesCross-references your packages against OSV.dev — returns only advisories that affect what you actually have installed

Your AI goes from guessing to knowing. In one tool call.


Supported ecosystems

EcosystemResolved lockfile (preferred)Fallback
RustCargo.lock—
npmpnpm-lock.yaml, yarn.lock, package-lock.json v1–v3package.json
Pythonpoetry.lock, uv.lockrequirements.txt (pinned ==)
Gogo.sumgo.mod

How it compares

context-snipeDependabotSnyksocket.dev
MCP native — AI gets the results directly✅❌❌❌
Reads resolved lockfiles (not just manifests)✅✅✅✅
100% local — nothing leaves your machine✅❌❌❌
No account, no signup, no API key✅❌❌❌
Binary size~1 MBN/A200 MB+N/A
Free, open source✅✅PartialPartial

CLI usage

context-snipe scan [PATH]    # vulnerability report (defaults to current dir)
context-snipe deps [PATH]    # list the full resolved dependency tree
context-snipe serve          # start the MCP server over stdio
context-snipe --help

A note on honesty

context-snipe tells you which vulnerable packages are present in your resolved dependency tree. It does not perform call-graph reachability analysis. Presence is not proof of exploitability — the vulnerable function may not be reachable in your code. The tool says so in its own output, by design.

No tool that runs in seconds can tell you a CVE is definitely not exploitable. We won't pretend otherwise.


How it works

  • MCP engine — hand-rolled JSON-RPC 2.0 over stdio. initialize, tools/list, tools/call, ping. stdout is the protocol channel; all diagnostics go to stderr.
  • Lockfile parsers — TOML for Cargo, JSON for npm, custom parsers for pnpm/yarn, line parsers for requirements.txt and Go modules.
  • OSV client — one querybatch call filters the full tree to packages with advisories, then a focused query per hit pulls details. CVSS v3.x base scores computed from vector strings. Duplicate advisories sharing a CVE are merged.
  • TLS via rustls — pure-Rust, no OpenSSL, no system crypto dependency. Works identically on Windows, macOS, and musl Linux.

Build from source

cargo build --release
# Binary at: target/release/context-snipe

Requires stable Rust. The release profile statically links the CRT — the binary is fully self-contained.


Roadmap

  • GitHub App — post CVE diffs on pull requests (shows what a PR introduces)
  • Policy layer — configurable CI failure thresholds per severity
  • More ecosystems — Ruby (Gemfile.lock), PHP (composer.lock), Java (pom.xml)

Contributing

PRs welcome. The codebase is ~1,000 lines of Rust split across:

src/
  main.rs    — CLI entry, mode routing
  mcp.rs     — JSON-RPC / MCP server
  deps.rs    — lockfile parsers
  osv.rs     — OSV.dev client + CVSS scoring
  scan.rs    — orchestration + report formatting
  http.rs    — ureq + rustls HTTP agent

Good first issues: adding a new lockfile format, improving CVSS display, adding output formats (JSON, SARIF).


License

MIT — free forever. No telemetry. No accounts. No cloud.

Ver o original no GitHub

Bom para

Relacionados