Skip to main content
Glama

MCPWorks

License: BSL 1.1 CI

Cut your AI agent token costs by 70-98%.

AI tool calls return entire datasets into the context window. A 500-row query becomes 47,000 tokens. MCPWorks runs your code in a secure sandbox instead — data stays in the sandbox, only the answer comes back.

Before:  AI calls tool → 500 records enter context → 47,000 tokens → AI summarizes → 200 token answer
After:   AI writes code → runs in sandbox → data never enters context → 300 tokens total

Self-host with docker compose up or use MCPWorks Cloud.

Quick Start

git clone https://github.com/MCPWorks-Technologies-Inc/mcpworks-api.git
cd mcpworks-api
cp .env.self-hosted.example .env
# Edit .env: set BASE_DOMAIN, generate ENCRYPTION_KEK_B64

mkdir -p keys
openssl ecparam -genkey -name prime256v1 -noout -out keys/private.pem
openssl ec -in keys/private.pem -pubout -out keys/public.pem

docker compose -f docker-compose.self-hosted.yml up -d

Health check: curl https://api.yourdomain.com/v1/health

Full guide: docs/SELF-HOSTING.md

Related MCP server: MCP of MCPs

How It Works

Claude / GPT / any LLM
    |
    | "from functions import query_leads; result = query_leads(tier='hot')"
    v
MCPWorks Sandbox (nsjail)
    |  Data queried, filtered, summarized inside sandbox
    |  Only the result exits
    v
{"hot_leads": 12, "top": "Acme Corp"}  ← 85 tokens, not 47,000

The AI writes Python or TypeScript. MCPWorks executes it in an isolated sandbox with access to your functions. The full dataset never enters the AI context window.

Features

Feature

What it does

Secure Sandbox

nsjail isolation: Linux namespaces, cgroups, seccomp. User code runs with zero privileges.

Token Efficiency

70-98% fewer tokens per operation. Data stays in sandbox, only results return.

Encrypted Credential Storage

Agent state stores API keys and secrets with AES-256-GCM envelope encryption. Injected server-side — never in the AI prompt.

Agent Runtime

Autonomous agents with scheduling, persistent state, webhooks, and AI orchestration. BYOAI — use any provider.

Function Hosting

Organize Python/TypeScript functions into services. Each namespace gets its own MCP endpoint.

Access Control

Per-agent function and state restrictions with glob patterns. Deny-takes-precedence.

Self-Hosted

docker compose up with bundled PostgreSQL, Redis, and Caddy. No external dependencies.

MCP Native

Full Model Context Protocol support. Works with Claude Desktop, Cursor, and any MCP client.

Stack

Python 3.11 / FastAPI / SQLAlchemy (async) / PostgreSQL / Redis / nsjail

Development

python3 -m venv venv && source venv/bin/activate
pip install -e ".[dev]"
docker compose up -d postgres redis
alembic upgrade head
uvicorn mcpworks_api.main:app --reload --port 8000
pytest tests/ -v

Project Structure

src/mcpworks_api/
    main.py           # FastAPI application
    mcp/              # MCP protocol handlers (create, run, agent)
    backends/         # Execution backends (nsjail sandbox)
    services/         # Business logic
    models/           # SQLAlchemy ORM models
    tasks/            # Agent orchestrator, scheduler
    middleware/       # Auth, rate limiting, routing
    core/             # Security, encryption, access control
    sandbox/          # Sandbox configuration, package registry

Community

Contributing

See CONTRIBUTING.md for development setup and PR process.

Security

Found a vulnerability? See SECURITY.md for responsible disclosure. Do not open public issues for security vulnerabilities.

License

Business Source License 1.1 — free for non-production use, production use for internal business purposes permitted. Converts to Apache 2.0 on 2030-03-22.

A
license - permissive license
-
quality - not tested
A
maintenance

Maintenance

Maintainers
1dResponse time
5dRelease cycle
4Releases (12mo)
Commit activity
Issues opened vs closed

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/MCPWorks-Technologies-Inc/mcpworks-api'

If you have feedback or need assistance with the MCP directory API, please join our Discord server