toolatlas-mcp 0.1.0

ToolAtlas — Discover, Govern, and Optimize MCP Tools

ToolAtlas-MCP

The Control Plane for Your MCP Tools

ToolAtlas sits between AI clients and MCP servers — giving you a central registry, intelligent proxy, per-tool governance, usage analytics, and a beautiful web dashboard.

pip install toolatlas-mcp
toolatlas start
# → http://localhost:8080

---

Why ToolAtlas?

MCP (Model Context Protocol) is exploding. Teams run 5, 10, 50 MCP servers — GitHub, Jira, Slack, Confluence, databases, APIs. Each exposes tools your AI agents can call. ToolAtlas is the missing layer that helps you see everything, control what matters, and understand usage.

---

Features

🔌 Universal MCP Proxy

Connect any number of MCP servers and expose them through a single proxy endpoint. Group tools by team, project, or purpose using proxies — each with its own tool set and settings.

🛡️ Per-Tool Governance

• Enable/disable tools per proxy — block dangerous operations without touching the server
• Custom descriptions — override tool descriptions so AI understands them in context
• Alias tools — rename tools per proxy for clarity

🏷️ Tool Registry & Discovery

• Auto-discover tools from any MCP server (SSE or stdio)
• Tag, domain, and annotate tools with business context
• Glossary terms — link tools to business concepts for richer descriptions

📊 Usage Analytics

• Every tool call is automatically tracked with duration, success/failure, and arguments
• Dashboard shows top tools, latency trends, and error rates
• Drill into individual call traces with full event timelines

🌐 Web Dashboard

Full React SPA for managing everything without touching the CLI:

• Servers, proxies, tools, glossary, analytics
• Edit tool metadata inline
• Real-time stats

---

Quick Start

1. Install

pip install toolatlas-mcp

2. Start

toolatlas start

Open http://localhost:8080 in your browser.

3. Add an MCP Server

# SSH-based MCP server
toolatlas server add my-github \
  --transport stdio \
  --command "npx" \
  --args "-y" "@modelcontextprotocol/server-github"

Or add SSE servers via the web UI.

4. Create a Proxy

toolatlas proxy add dev --description "Developer tools"
toolatlas proxy link dev my-github

5. Connect Your AI Client

Point your MCP client to:

http://localhost:8080/proxy/dev/message/{session_id}

---

Architecture

┌─────────────────┐     ┌─────────────────────────────────────┐     ┌─────────────────┐
│  MCP Clients    │     │         ToolAtlas                    │     │  MCP Servers     │
│                 │     │                                     │     │                  │
│  Claude         │────▶│  Proxy "dev"    ── governance ──▶ │────▶│  GitHub MCP      │
│  Cursor         │────▶│  Proxy "pm"     ── governance ──▶ │────▶│  Jira MCP        │
│  Custom Agents  │────▶│  Proxy "ops"    ── governance ──▶ │────▶│  Slack MCP       │
│                 │     │                                     │     │  Confluence MCP  │
└─────────────────┘     │  ┌─────────────┐  ┌─────────────┐ │     └─────────────────┘
                        │  │ Registry    │  │ Analytics   │ │
                        │  │ DB + API    │  │ Tracker     │ │
                        │  └─────────────┘  └─────────────┘ │
                        │  ┌──────────────────────────────┐  │
                        │  │  Web Dashboard (React SPA)   │  │
                        │  └──────────────────────────────┘  │
                        └─────────────────────────────────────┘

Clients speak MCP to ToolAtlas. ToolAtlas enforces governance, logs everything, and forwards to the real MCP servers.

---

CLI Reference

Command	Description
toolatlas start	Start the ToolAtlas server
toolatlas server add <name>	Register a new MCP server
toolatlas server list	List registered servers
toolatlas proxy add <slug>	Create a new proxy
toolatlas proxy list	List proxies
toolatlas proxy link <slug> <server>	Link a server to a proxy

---

Configuration

Set via environment variables with TOOLATLAS_ prefix:

Variable	Default	Description
TOOLATLAS_HOST	127.0.0.1	Bind address
TOOLATLAS_PORT	8080	HTTP port
TOOLATLAS_DATABASE_URL	sqlite+aiosqlite:///toolatlas.db	Database connection
TOOLATLAS_LOG_LEVEL	INFO	Log level

---

Development

git clone https://github.com/anomalyco/toolatlas-mcp
cd toolatlas-mcp

# Set up Python
pip install -e ".[dev]"

# Set up UI
cd ui && npm install && npm run build && cd ..

# Run tests
pytest

# Start dev server
toolatlas start --reload
# In another terminal: cd ui && npm run dev

---

License

MIT © ToolAtlas contributors