mpl

Quickstart

Two-minute start

Put mpl in front of an existing MCP or A2A server without touching your agents. Start by observing, then enforce once you trust the schemas.

  1. 1

    Install the proxy

    Install the mplx CLI, which bundles the sidecar proxy.

    cargo install mplx
  2. 2

    Run in transparent mode

    Point the proxy at your existing MCP (or A2A / HTTP) server. In transparent mode it observes and logs — it never blocks.

    mpl proxy http://your-mcp-server:8080
# dashboard at http://localhost:9080
  3. 3

    Learn schemas from live traffic

    Let the proxy watch real messages and propose stype contracts from what it sees.

    mpl schemas generate
  4. 4

    Approve the contracts

    Review the proposed schemas and approve the ones you trust.

    mpl schemas approve --all
  5. 5

    Switch to production (strict) mode

    Once you have seen the traffic, enforce. Invalid payloads and policy denials are now rejected at the proxy.

    mpl proxy http://your-mcp-server:8080 --mode production

Your agents don’t change. mpl observes, validates, and records what passes through. When you are ready to call it from code, reach for the Python or TypeScript SDK, or read the full documentation.

Frequently asked questions

What are the modes?

transparent observes and logs and never blocks (the default for onboarding); production / strict enforce — invalid payloads and policy denials are rejected at the proxy.

How do I call it from code?

Install a language SDK: pip install mpl-sdk (Python) or npm install @mpl/sdk (TypeScript). Both return valid, qom_passed (or qomPassed), and the response payload.

Do I need Kubernetes?

No — the proxy is a single binary you can run as a sidecar anywhere. A Helm chart (helm/mpl-proxy) is provided if you do deploy on Kubernetes.

Where does the audit trail go?

mpl produces the records; you own the sink. Point it at whatever append-only store your compliance program uses.