mcpd ¶

Run your agents, not your infrastructure.

mcpd is a toolchain and runtime developed by Mozilla AI that simplifies the configuration, execution and integration of Model Context Protocol (MCP) servers with your agentic application. It is intended to provide the same experience across local, development and production environments.

It removes the friction of cross-language server orchestration, secrets management, version pinning, and lifecycle control.

Why `mcpd`?¶

Traditional agent frameworks often embed complex subprocess logic, brittle startup scripts, and ad-hoc 'desktop style' config.

mcpd replaces this with:

Zero-Config Tool Setup
No cloning repos or manual per-server setup. Install the runtime your chosen server uses, then mcpd add and mcpd daemon handle the rest.

Language-Agnostic Tooling
Use MCP servers written in Python (uvx), JavaScript/TypeScript (npx) in your code via a HTTP REST API that supports routing to MCP Servers.

Declarative Tool Management
Define version-pinned MCP servers and tools in .mcpd.toml. Reproducible, consistent, and CI-friendly.

Project config separated from runtime variables
Exportable args and environment variables per server e.g. ~/.config/mcpd/secrets.dev.toml. Never commit dev specific vars to Git again.

Unified Dev Experience
One command: mcpd daemon. Starts and manages all servers behind the scenes.

Intuitive SDK Integration
The Python mcpd_sdk makes calling tools feel like native function calls; no HTTP, STDIO, or SSE boilerplate. Even easier for users of any-agent via .agent_tools().

Seamless Local-to-Prod Transition
The same .mcpd.toml and agent code work in dev, CI, and cloud environments without modification.

Built for Dev & Infra¶

Feature	Dev Workflow	Production Benefit
`.mcpd.toml`	Version-controlled agent tool spec	Declarative IaC for tools
`mcpd daemon`	Run everything locally	Run in container alongside your agentic app
`mcpd config export`	Discover required vars	Populate CI/CD pipelines
Secure secrets store	Local overrides per dev	Injected via Control Plane
Same binary	Local builds	Cloud deployment
SDKs for Python	Iterate locally	Plug into prod orchestrators

Deploy Anywhere¶

mcpd is runtime-flexible and infrastructure-agnostic:

⚙️ Works in any container or host where the runtimes required by your configured servers are available (uv, npx, and/or Docker)
☁️ Multi-cloud ready (AWS, GCP, Azure, on-prem)
♻️ Low resource overhead via in-process server management

📦 Install¶

Homebrew¶

Add the Mozilla.ai tap:

brew tap mozilla-ai/tap

Then install mcpd:

brew install mcpd

Or install directly from the cask in a single command:

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

Installation methods

Please see our Installation page for additional ways to install and run mcpd.

Runtime prerequisites

Installing the mcpd binary is separate from installing MCP server runtimes. Install only the runtime(s) your configured servers use: uv for uvx::..., npx for npx::..., and Docker for docker::.... The tutorial and most examples on this site use uvx, so uv is the most common first requirement.

📚 Explore the Docs¶

Use the sidebar to explore:

About Mozilla.ai¶

This project is built and maintained by Mozilla.ai, a mission-driven organization reimagining AI for the public good.

Have ideas or feedback? Contributions welcome via GitHub.