Run your agents, not your infrastructure.

Built by Mozilla AI

📚 mcpd official docs

---

mcpd is a daemon that manages your MCP servers via declarative configuration, exposing them as clean HTTP endpoints. This bridges the gap between your agents and your infrastructure, handling the messy work of lifecycle management, secret injection, and environment promotion so you don't have to.

⚙️ How it Works

Under the hood, mcpd spawns MCP servers as STDIO subprocesses and proxies requests over HTTP.

🚀 Quick Start

Prerequisites

You need the following installed before running mcpd:

• Docker
• npx (Node.js package runner)
• uvx (Universal virtual environment manager)

Installation

via Homebrew

(Works for both macOS and Linux)

Add the Mozilla.ai tap:

brew tap mozilla-ai/tap

Then install mcpd:

brew install mcpd

Or install directly from the cask:

brew install --cask mozilla-ai/tap/mcpd

Please read our docs to install via GitHub releases or local Go Binary build.

Using mcpd

# Initialize a new project and create a new .mcpd.toml file
mcpd init

# Add an MCP server to .mcpd.toml
mcpd add time

# Set the local timezone for the MCP server
mcpd config args set time -- --local-timezone=Europe/London

# Start the daemon in dev mode with debug logging
mcpd daemon --dev --log-level=DEBUG --log-path=$(pwd)/mcpd.log

Now that the daemon is running, let's call the get_current_time tool provided by the time MCP server

# Check the time
curl -s --request POST \
  --url http://localhost:8090/api/v1/servers/time/tools/get_current_time \
  --header 'Accept: application/json, application/problem+json' \
  --header 'Content-Type: application/json' \
  --data '{
  "timezone": "Europe/Warsaw"
}'

API docs will be available at http://localhost:8090/docs.

💡 Why mcpd?

Engineering teams build agents that work locally, then struggle to make them production-ready across environments. mcpd bridges this gap with declarative configuration and secure secrets management.

• Declarative & reproducible – .mcpd.toml defines your tool infrastructure
• Language-agnostic – Python, JS, Docker containers via unified HTTP API
• Dev-to-prod ready – Same config works locally and in containers

🏗️ Built for Dev & Production

Development Workflow	Production Benefit
mcpd daemon runs everything locally	Same daemon runs in containers
.mcpd.toml version-controlled configs	Declarative infrastructure as code
Local secrets in ~/.config/mcpd/	Secure secrets injection via control plane
mcpd config export exports version-control safe snapshot of local configuration	Sanitized secrets config and templates for CI/CD pipelines

📦 SDKs

mcpd SDKs

Language	Repository	Status
Python	mcpd-sdk-python	✅
JavaScript	mcpd-sdk-javascript	✅

mcpd plugin SDKs

Plugin SDKs are built using the mcpd plugin Protocol Buffers specification.

Language	Repository	Status
Go	mcpd-plugins-sdk-go	✅
.NET	mcpd-plugins-sdk-dotnet	✅
Python	mcpd-plugins-sdk-python	✅

💻 Development

If you are developing mcpd, you will need:

• Go (v1.25.6+ recommended)

Build local code:

make build

Run tests:

make test

Validate Mozilla AI registry (when modifying registry files):

make validate-registry

Run the local documentation site (requires uv), dynamically generates command line documentation:

make docs

🧩 The Mozilla.ai Stack

mcpd is the "Tooling Layer" of the Mozilla.ai ecosystem. These tools are designed to work together or standalone.

Layer	Tool	Function
Compute	llamafile	Local LLM inference server
Interface	any-llm	Unified Python library for LLM inference
Logic	any-agent	Orchestration and agent loops
Tools	mcpd	(You are here) Tool sandbox & router
Safety	any-guardrail	Input/Output validation

🤝 Contributing

Please see our Contributing to mcpd guide for more information.

📄 License

Licensed under the Apache License 2.0.