Skip to content

Overview

Enterprise-grade MCP servers, governed by default.

Pontifex adds authentication, least-privilege scopes, and a full audit trail to every tool call. It builds on the official MCP Python SDK.

Quickstart

Why Pontifex

Your APIs and internal systems were built for your applications and employees, not for an autonomous agent. Point an agent at them and your security team asks four questions:

Who is calling, what can they touch, how often, and what did they do?

MCP, the protocol agents use to call tools, standardizes the connection and leaves the control to you. So your team builds a strong pilot, and it stalls at the first system that holds real data.

Pontifex answers those four questions, so you can ship the pilot.

What Pontifex does

Pontifex wraps your existing APIs and services in governed MCP tools. It authenticates every request and writes an audit record before any handler runs. You ship to production; your data stays in your environment.

flowchart TB
    agent["AI agent<br/>Claude Desktop, your agents"]

    subgraph pontifex ["pontifex-mcp&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;"]
        gate["Authenticate · check scope · rate limit"]
        tool["Your tool"]
        audit["Audit log"]
        gate --> tool
        tool --> audit
    end

    agent -->|MCP| gate
    tool --> adapters["Data adapters"]
    adapters --> systems["Your systems<br/>APIs, databases, services"]

What it changes

Without a governance layer With Pontifex
An agent's access is all-or-nothing You scope each caller to the exact tools they need
"Trust the agent" You trust a verified identity on every call
No record of what the agent did A full audit row per call: caller, tool, data source, latency
One slow upstream stalls everything Rate limiting, failover, and circuit breaking contain it
Your data flows through a vendor You self-host; nothing leaves your environment

What you get

  • Nothing runs unauthenticated

    Every call carries a verified identity: an OAuth 2.1 JWT or an sk_… API key. Pontifex checks it against any OIDC provider before your handler runs.

  • Least privilege, enforced

    Scopes are namespace:resource:action, declared per tool. A caller cannot widen their own access at runtime.

  • Audit you can hand to a reviewer

    Pontifex records every call: the caller, the tool, the parameters, the data source, and the latency.

  • Resilient under load

    Per-caller rate limiting, source failover, and circuit breaking keep one slow upstream from stalling the server.

  • No code for an existing API

    Point a config file at an OpenAPI spec, and Pontifex governs every allowlisted operation as a tool. Connectors

  • Yours to run

    A Python library you self-host. No third party sits in the request path. Apache-2.0 licensed.

  • Coding-agent friendly

    Your coding agent reads an official skill bundled with the package, so it builds with pontifex-mcp on guidance that matches the version you installed. Use with your coding agent

Pontifex vs. the MCP SDK alone

The SDK gives you a server. Pontifex makes it safe to run on real systems.

MCP Python SDK Pontifex MCP
Define and serve tools ✅ ✅ (built on it)
Authenticate callers — ✅ API keys + OAuth 2.1
Per-caller scopes — ✅ namespace:resource:action
Audit log — ✅ every call, to Postgres
Rate limiting — ✅ per caller
Resilience (failover, breakers) — ✅
Onboard an OpenAPI API with no code — ✅

Built on open standards

Pontifex does not ask you to bet on a platform. It builds on the official MCP Python SDK and uses OAuth 2.1 and standard JWTs for identity, so you bring any OIDC provider: Auth0, Entra, Clerk, or Keycloak. It reads OpenAPI to onboard existing systems and speaks RFC 9728 and RFC 8693 for discovery and token exchange.

Pair it with any AI vendor and run it anywhere you run Python. Drop the dependency to remove it; your tools stay standard MCP underneath.

You hold the data

Pontifex is a library you run, not a service you send data to.

It sits inside your environment, between the agent and your systems. No third party sits in the request path. You supply the database and credentials from your own infrastructure. Pontifex hardcodes nothing and phones nothing home.

That is what lets your security and compliance teams sign off on "we are connecting AI to customer data."

Where to next

  • Building with it?

    An authenticated, audited server running in minutes.

    Quickstart

  • Reviewing the security?

    The model behind "safe to point at production."

    Security model

  • Onboarding an existing API?

    Governed tools from an OpenAPI spec, no handler code.

    Connect an API

  • Sizing the architecture?

    How a request travels through auth, scopes, and audit.

    How a request flows

Apache-2.0 licensed.