convez.ai
MCP Server
major-matters

mcp-scanner

major-matters/mcp-scanner

Ver no GitHub
InicialPluginScore · de 0 a 100

Open scan of the public MCP server ecosystem — discovery, maintenance, transport, and licensing signals behind Major Labs' State of MCP writeups. Read-only (GitHub API). v0.

Instalação

$npx convez-ai major-matters/mcp-scanner

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
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 2 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

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

mcp-scanner · Sentinel (v0)

An open scan of the public Model Context Protocol (MCP) server ecosystem. It pulls MCP server repositories from GitHub, captures observable signals (transport mode, maintenance, language, license, manifest presence), and stores them in SQLite. The resulting dataset is the factual basis for Major Labs' State of MCP writeups.

v0 scope: discovery and metadata only. Everything here is read-only and uses the public GitHub API (via the gh CLI). It does not connect to, probe, or send anything to any MCP server. Active vulnerability probing (SSRF, injection, etc.) is future work and is intentionally not in this repository.

What it does

  1. discover.py — finds candidate MCP server repos via GitHub topic/org/name search (topic:mcp-server, topic:model-context-protocol, the modelcontextprotocol org, etc.), deduped by owner/repo.
  2. analyze.py — for each repo, reads the README and dependency manifests (package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod) via the GitHub API and classifies the likely transport mode (stdio / HTTP / both).
  3. security.py — READ-ONLY static security sweep: fetches public source for catalogued repos via the GitHub API and flags security-relevant code patterns (command injection, SSRF surface, eval/exec, unsafe deserialization, hardcoded secrets). It never connects to, runs, or probes a server. Findings are heuristic static signals, tuned for precision. Only aggregate statistics are published (see majorlabs.co/security); per-repo findings stay in the local database for coordinated maintainer disclosure.
  4. report.py — prints population, maintenance, transport, and licensing summaries.
  5. export_dataset.py — writes the published dataset (see below).

Run

gh auth login                 # uses your own GitHub token; nothing is embedded
python3 discover.py
python3 analyze.py
python3 report.py
python3 export_dataset.py      # writes dataset/*.csv

Output: scans.db (local SQLite, gitignored) and dataset/*.csv.

The dataset (dataset/)

  • mcp-servers.csv — one row per repo: name, owner, description, stars, forks, language, license, archived/fork flags, created/pushed/updated dates, topics, the transport classification, and which dependency manifests are present.
  • discovery-runs.csv — the exact search queries and result counts, for methodology transparency.

Deliberately excluded: the raw README and manifest text bodies that analyze.py stores locally in scans.db. They carry no analytical value for the reports and are the only place where author emails or example credentials from other people's repositories could appear, so the published dataset omits them. Re-run discover.py + analyze.py to rebuild the full local database yourself.

Methodology and limitations

This is a best-effort, reproducible scan, not a definitive census. Known caveats:

  • Transport classification is heuristic (regex over README phrasing), so "stdio / HTTP / both / unclear" is a signal, not ground truth.
  • Discovery is search-driven, so repos that are not tagged or named conventionally are missed. discovery-runs.csv records exactly what was queried.
  • Counts reflect the GitHub state at scan time (discovered_at / analyzed_at).

Ethics

  • Public data only (GitHub API, public repositories). No authentication to, or probing of, third-party servers.
  • Respects GitHub API rate limits.
  • For any vulnerability later surfaced at a named maintainer's project, we follow coordinated disclosure (30 days before publication).

License

MIT.

Ver o original no GitHub

Bom para

Relacionados