convez.ai
MCP ServerSegurançaIntermediário
AgentSafe-AI

ToolTrust Directory

AgentSafe-AI/tooltrust-directory

Registro seguro de MCP servers com scoring A-F de risco

Ver no GitHub
BomPluginScore · de 0 a 100

Registro público e continuamente atualizado de ferramentas para agentes de IA, com avaliação automática de riscos de segurança (injeção de prompt, exfiltração de dados, escalação de privilégio). Fornece classificação A-F para mais de 1.200 MCP servers e plugins, com detecção de incidentes em cadeias de suprimento.

Instalação

$npx convez-ai AgentSafe-AI/tooltrust-directory

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
7
Forks
3
Acompanhando
1
Saúde
62%
Último commit
há 1 dia
Versão
—
Issues abertas
6
Contribuidores
3

Casos de uso

  • Validar segurança de MCP servers antes de integração
  • Pesquisar vulnerabilidades conhecidas em ferramentas de IA
  • Avaliar risco de supply-chain de dependências npm em agents

Veredito

Recurso essencial para equipes que deployam agentes em produção. O scoring contínuo e a cobertura de incidentes recentes (LiteLLM, axios malicioso) reduzem riscos operacionais reais, embora a confiabilidade dependa da qualidade do scanner subjacente.

README

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

🛡️ ToolTrust Directory

This repo hosts tooltrust.dev — the website and pre-scanned report data. If you want to scan your own MCP servers, go to tooltrust-scanner.

A public registry of AI agent tools, continuously scanned for prompt injection, data exfiltration, and privilege escalation by ToolTrust Scanner.

🚨 Supply-Chain Incident Coverage (March 2026) ToolTrust now detects and blocks confirmed supply-chain incidents including the LiteLLM / TeamPCP compromise and the malicious axios npm publish (axios@1.14.1, axios@0.30.4). For npm-backed MCP servers, ToolTrust also scores dependency visibility, transitive lockfile evidence, lifecycle scripts, and IOC indicators such as plain-crypto-js.

ToolTrust Directory UI

Tools Audited Last Scan License: MIT Schema


📊 Security Registry

Top 50 by popularity. View all 1246 tools → Full Directory · data/reports/ · docs/tools/

ToolVersionPopularityGradeKey FindingsScanned
playwright-mcp0.0.7614.5M/moD📦 AS-004 ×23, 🔑 AS-002 ×15, ⚡ AS-006 ×2, ⚡ AS-011 ×6Jun 11
chrome-devtools-mcpchrome-dev…9.3M/moD📦 AS-004 ×29, 🔑 AS-002 ×16, ⚡ AS-011 ×4, ⚡ AS-006Jun 11
upstash-context7-mcp1.0.305.0M/moBAS-014 ×2, 🔑 AS-002, ⚡ AS-011Jun 11
context71.0.305.0M/moBAS-014 ×2, 🔑 AS-002, ⚡ AS-011Jun 11
gemini-cli0.47.0-pre…3.1M/moCAS-014 ×56, 🔑 AS-002 ×35, ⚡ AS-011 ×11Jun 11
cloudflare-containers0.3.21.4M/moA🔑 AS-002 ×5, ⚡ AS-011, AS-014 ×7May 30
mcp-server-filesystemtypescript…1.3M/moC🔑 AS-002 ×15, 📦 AS-004 ×14, ⚡ AS-011Jun 11
mcp-server-githubtypescript…563.6k/moC🔑 AS-002 ×35, 📦 AS-004 ×26, ⚡ AS-011 ×18Jun 11
xcodebuildmcp2.6.2532.0k/moBAS-014 ×71, 🔑 AS-002 ×35, ⚡ AS-011 ×3Jun 11
n8n-mcp2.57.3528.8k/moC📦 AS-004 ×7, 🔑 AS-002 ×7, ⚡ AS-011 ×2Jun 11
mcp-server-sequential-thinkingtypescript…438.7k/moB📦 AS-004Jun 11
firecrawl-mcp-server3.2.1377.0k/moC🔑 AS-002 ×14, AS-014 ×8, ⚡ AS-011 ×7Jun 11
cameroncooke-xcodebuildmcp2.3.2329.3k/moBAS-014 ×71, 🔑 AS-002 ×35, ⚡ AS-011 ×3May 30
notion-mcp-server2.1.0316.9k/moD🔑 AS-002 ×30, 📦 AS-004 ×22, ⚡ AS-011 ×22Jun 11
figma-context-mcp0.12.0296.4k/moBAS-014 ×9, 🔑 AS-002, ⚡ AS-011Jun 11
tavily-mcp0.2.20165.1k/moC🔑 AS-002 ×10, 📦 AS-004 ×5, ⚡ AS-011 ×5Jun 11
tavily-ai-tavily-mcp0.2.19155.7k/moC🔑 AS-002 ×10, 📦 AS-004 ×5, ⚡ AS-011 ×5May 30
ruflo3.10.41147.4k/moBAS-014 ×33, 🔑 AS-002 ×25, ⚡ AS-011Jun 11
context-mode1.0.162131.5k/moC🔑 AS-002 ×9, AS-014 ×7, ⚡ AS-011Jun 11
mcp-server-kubernetes3.9.0104.2k/moBAS-014 ×22, 🔑 AS-002 ×6, ⚡ AS-011 ×3Jun 11
mcp-server-brave-searchtypescript…104.1k/moC🔑 AS-002 ×12, ⚡ AS-011 ×7, AS-014 ×8, 🗝️ AS-010 ×2Jun 11
circleci-public-mcp-server-circleci0.15.1101.4k/moD🔑 AS-002 ×21, 📦 AS-004 ×17, ⚡ AS-011 ×13, 📐 AS-003 ×2May 30
ms-365-mcp-server0.118.197.9k/moD📦 AS-004 ×178, 🔑 AS-002 ×377, ⚡ AS-011 ×172Jun 11
mcp-server-timetypescript…87.0kAAS-014 ×2Jun 11
apify-mcp-server0.10.1384.8k/moD🔑 AS-002 ×27, ⚡ AS-011 ×7, AS-014 ×16, ⚡ AS-006 ×2Jun 11
mobile-mcp0.0.5977.5k/moC📦 AS-004 ×23, 🔑 AS-002 ×5, ⚡ AS-011Jun 11
exa-mcp-server3.2.167.9k/moC🔑 AS-002 ×4, 📦 AS-004 ×2, ⚡ AS-011 ×2Jun 11
desktopcommandermcp0.2.4266.2k/moC🔑 AS-002 ×22, AS-014 ×26, ⚡ AS-011 ×8, 📐 AS-003Jun 11
brave-search-mcp-server2.0.8366.1k/moC🔑 AS-002 ×12, ⚡ AS-011 ×7, AS-014 ×8, 🗝️ AS-010 ×2Jun 11
claude-task-master0.20.064.2k/moBAS-014 ×14, 🔑 AS-002 ×9, ⚡ AS-011Jun 11
railway-mcp-server0.1.1146.5k/moCAS-012, 🔑 AS-002 ×57, AS-014 ×36, ⚡ AS-011 ×13, 🗝️ AS-010May 30
obsidian-mcp-server3.2.442.3k/moB🔑 AS-002 ×9, AS-014 ×13, ⚡ AS-011 ×2Jun 11
dive0.14.238.7k/moC🔑 AS-002 ×3, ⚡ AS-011 ×2, AS-014 ×2Jun 11
git-mcp-server2.15.135.5k/moC🔑 AS-002 ×39, 📦 AS-004 ×28, ⚡ AS-011 ×9Jun 11
tacticlaunch-mcp-linear1.1.231.6k/moCAS-014 ×42, 🔑 AS-002 ×21, ⚡ AS-011 ×8May 30
github-mcp-server1.2.030.6kC🔑 AS-002 ×75, ⚡ AS-011 ×36, AS-014 ×86, 📐 AS-003, 🗝️ AS-010Jun 11
brightdata-mcp2.10.030.5k/moC🔑 AS-002 ×67, ⚡ AS-011 ×58, AS-014 ×65Jun 11
aas-ee-open-websearch2.1.1128.5k/moC🔑 AS-002 ×7, 📦 AS-004 ×6, ⚡ AS-011 ×6Jun 11
antvis-mcp-server-chart0.9.1026.1k/moBAS-014 ×26, 🔑 AS-002, ⚡ AS-011May 30
mcp-server-chart0.9.1025.8k/moBAS-014 ×26, 🔑 AS-002, ⚡ AS-011Jun 11
neo11.19.122.3k/moBAS-014 ×8, 🔑 AS-002, ⚡ AS-011Jun 11
kastalien-research-thoughtbox1.2.020.1k/moCAS-014 ×3, 🔑 AS-002 ×3, ⚡ AS-011May 30
kastalien-research-clear-thought-two1.2.020.1k/moCAS-014 ×3, 🔑 AS-002 ×3, ⚡ AS-011May 30
mcp-server-asana1.6.019.6k/moC🔑 AS-002 ×8, ⚡ AS-011 ×3, AS-014 ×10Jun 11
postman-mcp-server2.9.019.1k/moC🔑 AS-002 ×53, ⚡ AS-011 ×15, AS-014 ×41Jun 11
dainfernalcoder-perplexity-mcpFeature17.5k/moB🔑 AS-002, ⚡ AS-011, AS-014May 30
openmetadata1.2.114.2kCAS-014 ×31, 🔑 AS-002 ×11, ⚡ AS-011 ×8Jun 11
xiaohongshu-mcp2026.05.28…14.1kC🔑 AS-002 ×7, ⚡ AS-011 ×5, AS-014 ×10Jun 11
skill-seekers3.7.014.0kBAS-012, 🔑 AS-002 ×2, ⚡ AS-011 ×2, AS-014 ×2Jun 11
mcp-server-cloudflareworkers-ob…12.8k/moD🔑 AS-002 ×5, ⚡ AS-011 ×2, AS-014 ×2, ⚡ AS-006Jun 11

⚖️ Grading System

GradeGateway ActionDescription
S 🌟ALLOWReserved for dynamic analysis
AALLOWMinimal risk. Safe for production agents.
BALLOW + rate limitLow risk. Minor issues, but generally safe.
CREQUIRE_APPROVALModerate risk. Remediation recommended.
DREQUIRE_APPROVALHigh risk. Use only in isolated environments.
FBLOCKCritical risk. Do not use in agentic pipelines.

Full methodology: docs/methodology.md


🔍 Check Catalog

ToolTrust Scanner check IDs referenced in all reports:

IDSeverityDetects
🛡️ AS‑001CriticalTool Poisoning — Adversarial prompts hidden in tool descriptions (ignore previous instructions, <INST>)
🔑 AS‑002High/LowPermission Surface — exec, network, db, fs beyond stated purpose; over-broad input schema
📐 AS‑003HighScope Mismatch — Tool name contradicts its permissions (e.g. read_config with exec)
📦 AS‑004High/CriticalSupply Chain CVEs — Known CVEs in bundled dependencies via OSV
🔓 AS‑005HighPrivilege Escalation — admin/:write OAuth scopes; sudo/impersonate in descriptions
⚡ AS‑006CriticalArbitrary Code Execution — evaluate_script, _evaluate suffix, execute javascript, page.evaluate() patterns
ℹ️ AS‑007InfoInsufficient Tool Data — Tool lacks a valid description or schema
🚨 AS‑008CriticalKnown Compromised Package — Offline embedded blacklist of confirmed supply-chain attacks (LiteLLM 1.82.7/1.82.8, Trivy v0.69.4-v0.69.6, Langflow <1.9.0, Axios 1.14.1/0.30.4). Zero-latency, no network required.
🔤 AS‑009MediumTyposquatting — Tool name within edit-distance 2 of a well-known MCP tool, suggesting impersonation
🗝️ AS‑010MediumSecret Handling — Input params accepting API keys/passwords; credentials logged insecurely
⚡ AS‑011LowDoS Resilience — No rate-limit, timeout, or retry config on network/exec tools
🔄 AS‑012HighRug-Pull — Tool set changed between scans of the same version without a version bump (directory pipeline only)
👥 AS‑013High/MediumTool Shadowing — Duplicate or near-duplicate tool name hijacks calls intended for a trusted tool
ℹ️ AS‑014InfoDependency Inventory Unavailable — MCP server exposed neither metadata.dependencies nor a repo_url, so supply-chain coverage is limited and must be treated as incomplete
⚠️ AS‑015Medium/HighSuspicious NPM Lifecycle Script — npm dependency publishes preinstall / postinstall / similar install-time scripts; severity rises for remote-fetch or inline-execution patterns
🚨 AS‑016CriticalSuspicious NPM IOC Dependency — published npm metadata or install-time scripts reference a known malicious IOC package, domain, URL, or reviewed script pattern such as plain-crypto-js, even if the top-level package name is new
⚠️ AS‑017MediumSuspicious Data Exfiltration Description — tool description explicitly suggests sending user data, content, or conversation history to external / remote endpoints, without classifying it as prompt injection
ℹ️ AS‑018InfoEmbedded MCP Server Detected — source-level MCP SDK usage was found, but tools could not be enumerated from a manifest or live handshake, so manual review is still required
🔓 AS‑019HighUnauthenticated MCP Route Exposure — embedded MCP HTTP routes expose the same handler without equivalent authentication middleware

Full details → docs/methodology.md


🤖 AI Agent Integration

Let your AI agent scan its own tools. Add ToolTrust as an MCP server in your .mcp.json or claude_desktop_config.json:

{
  "mcpServers": {
    "tooltrust": {
      "command": "npx",
      "args": ["-y", "tooltrust-mcp"]
    }
  }
}

This gives your agent five security tools:

ToolDescription
tooltrust_scan_configScan all MCP servers in your .mcp.json or ~/.claude.json in parallel
tooltrust_scan_serverLaunch and scan a specific MCP server
tooltrust_scanner_scanScan a JSON blob of tool definitions
tooltrust_lookupLook up a server's trust grade from this directory
tooltrust_list_rulesList all security rules with IDs and descriptions

Claude Code users: ask your agent to run tooltrust_scan_config to audit every MCP server in your project in one shot.


🤝 Contribute

Request a scan — open an issue with the tool's public URL and version.

Dispute a finding — open an issue referencing the finding ID (e.g. AS-002).

Integrate ToolTrust Scanner — see docs/dev.md for the data pipeline and schema spec.


📛 Add to your README

If your MCP server was audited and earned a grade, add our badge to your repo:

Grade A (recommended) — copy this into your README:

[![ToolTrust Grade A](https://raw.githubusercontent.com/AgentSafe-AI/tooltrust-directory/main/docs/badges/grade-a.svg)](https://github.com/AgentSafe-AI/tooltrust-directory)

Other grades — replace grade-a with grade-s, grade-b, grade-c, grade-d, or grade-f:

GradeBadge
SGrade S
AGrade A
BGrade B
CGrade C
DGrade D
FGrade F

Badges link to this directory. Generate SVGs locally: go run ./cmd/badge


⚙️ Automation

The registry table above is kept up to date by a daily GitHub Actions workflow:

.github/workflows/daily-audit.yml   ← cron 00:00 UTC + manual dispatch

Each run:

  1. Discovers popular MCP servers via GitHub Search (50+ stars) plus Smithery-native servers (10+ uses)
  2. Scans new/updated tools with ToolTrust Scanner + OSV supply-chain analysis
  3. Publishes updated reports to data/reports/ and regenerates this README

Licensed MIT. Scanner engine: ToolTrust Scanner.

Ver o original no GitHub

Bom para

Relacionados