🏴‍☠️ AgentVault

The essential toolkit for AI agents — 9 tools in one install, zero API keys.

🌐 Website · 📦 npm · 📋 JSR · 📊 Ecosystem Report

⚡ 30-Second Start

npx agentvault-tools

That's it. Your agent now has web search, persistent memory, calculator, and more — no API keys, no config, no cost.

Claude Desktop

{
  "mcpServers": {
    "agentvault": {
      "command": "npx",
      "args": ["-y", "agentvault-tools"]
    }
  }
}

Cursor / Windsurf

{
  "mcpServers": {
    "agentvault": {
      "command": "npx",
      "args": ["-y", "agentvault-tools"]
    }
  }
}

🛠️ The Starter Kit — 9 Tools in One Package

agentvault-tools is the first package every AI agent should install:

Memory is 100% local — your data stays on your machine. SQLite + FTS5, zero cloud dependency.

📦 All 7 Packages

📊 MCP Ecosystem Health

We patrol the MCP ecosystem daily with automated agents. Key findings:

• 57% of MCP servers on npm fail to connect (26 of 46 tested)
• All failures produce the same error: Connection closed
• Many packages are published but broken or require undocumented setup
• 21 servers successfully scored, average quality: 89/100
• 295 tools catalogued across operational servers

📋 Full Ecosystem Report →

Top Scored Servers

🔒 Security

Tool Poisoning Detection

import { createGuard } from 'agentvault-guard';

const guard = createGuard({ mode: 'strict' });
const result = guard.check(toolName, args);
if (!result.allowed) {
    console.error('🔴 Blocked:', result.reason);
}

Resilient Calls

import { withRetry } from 'agentvault-retry';

const result = await withRetry(() => callTool('search', { query: 'test' }), {
    maxRetries: 3,
    timeoutMs: 5000,
});

Cache Results

import { createCache } from 'agentvault-cache';

const cache = createCache({ ttlMs: 60000 });
const result = await cache.wrap('search:test', () => callTool('search', { query: 'test' }));

🤖 Framework Integration

CrewAI

from crewai import Agent
from crewai.mcp import MCPServerStdio

agent = Agent(
    role="Research Analyst",
    goal="Gather data securely via audited MCP tools",
    mcps=[
        MCPServerStdio(
            command="npx",
            args=["-y", "agentvault-tools"],
        )
    ]
)

Audited Proxy (for any server)

{
  "mcpServers": {
    "github-audited": {
      "command": "npx",
      "args": ["-y", "agentvault-gateway", "--target", "npx -y @modelcontextprotocol/server-github"]
    }
  }
}

Architecture

┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│  AI Agent    │────▶│  Gateway     │────▶│  MCP Server  │
│  (Claude,    │     │  (Proxy +    │     │  (Any npm    │
│   Cursor,    │     │   Telemetry) │     │   MCP pkg)   │
│   etc.)      │     └──────┬───────┘     └──────────────┘
└──────────────┘            │
                            ▼
                    ┌──────────────┐     ┌──────────────┐
                    │  Telemetry   │     │  AgentScore   │
                    │  Data Lake   │     │  Scan Data    │
                    └──────┬───────┘     └──────┬───────┘
                           │                     │
                           ▼                     ▼
                    ┌─────────────────────────────┐
                    │      Intelligence Engine     │
                    │   Quality Reports, Alerts,   │
                    │   Trust Scores, Rankings     │
                    └─────────────────────────────┘

License

MIT — Built by AgentVault 🏴‍☠️