Skip to content
OuterSpace Apizr: a space explorer in an orange suit in front of a yellow spacecraft

Connect selected Python functions to AI agents through MCP, or expose them as REST APIs.

Apizr is an open-source capability compiler. Analyze existing Python, choose the public functions, generate their interfaces and define how calls execute. Your business logic stays in Python.

Quickstart: MCP and REST calls The full journey

Latest published stable: 0.3.0 · Released 22 September 2026

Preparing 0.4 — not released. Use the stable Quickstart above for your first server. The 0.4 development guide covers the compiler API, project files, remote Git, isolated plugins and OCI/Attest delivery; those features are not in stable 0.3.0. Python 3.11–3.14 · GPL-3.0-or-later.

From Python code to REST and MCP

Bring existing Python expertise into your applications and AI workflows. See how Apizr turns a route optimizer into callable interfaces.

Your expertise. Ready to connect. Apizr demonstration. Watch the demo

English narration · English captions

Read the transcript

Put your business logic to work. For an operations team, time on the road matters. Your Python code already plans better routes. Apizr helps make that expertise available to the people and systems that need it.

Keep the expertise you already have. Here, a specialist has prepared a route optimizer in a notebook. The next step is to connect it to the dispatch application, using the same calculation.

Understand what you can expose. First, inspect the notebook with Apizr. It identifies the functions and reports which can generate an interface. Your team chooses the capability to make available.

Connect your operational applications. Select the route optimizer and generate REST. Apizr creates the application and its API contract. The dispatch system now has an interface it can call.

Generated code your team can inspect. These are the generated files. Your developers can inspect and run them. The adapter connects incoming requests to your existing function, keeping the business calculation in one place.

Put the generated API to work. Start the generated application and send a list of stops. The response contains a proposed route and its driving estimate. This is what the dispatch interface will use.

See the operational impact. Now the result becomes visible. On this eight-stop example, the estimated driving time drops from about one hundred and one minutes to forty eight.

Adapt when the workload changes. For a different twelve-stop tour, call the same API again. The route changes, with a fifty four percent improvement over this starting order. These estimates do not include live traffic.

Make the same capability available to AI. Next, use Apizr to generate an MCP server from the same function. An assistant can use the existing calculation as a tool, rather than estimate the route itself.

Turn a business request into an action. Ask Ollama to add the Pantheon while keeping the existing stops. Our local interface passes the generated tool definition to the model, then forwards its chosen call through MCP.

The same logic serves another channel. The tool returns a thirteen-stop tour. The map updates from that result. The application and the assistant now use the same underlying business logic.

Existing expertise. New applications. Inspect your Python. Choose what to expose. Generate REST and MCP interfaces with Apizr. Bring existing expertise into your applications and AI workflows.

Choose what becomes public

Apizr scans code without executing it. Readiness is its assessment of whether the available evidence supports an interface. Your exposure policy explicitly selects the public functions: ready does not mean exposed.

A bundle contains the generated server, its contracts and the source needed by the selected functions. Support helpers can be included without becoming public. Start with MCP tools or REST endpoints, then explore selection and policies.

Decide how calls execute

Generating a bundle is static analysis. Running that bundle executes your code. Use trusted source and dependencies. The Quickstart uses direct mode: functions run inside the server process, and no Docker is needed.

For governed execution, an execution policy chooses a fresh process or container per call and the required limits. Local workers provide bounds and cleanup, but do not isolate files or network access. OCI adds reviewed container controls, not a VM or an untrusted-code guarantee. See execution boundaries and the optional strict OCI profile.

Find the right server and guide

The generated MCP server serves your selected Python functions. The separate Apizr analysis MCP server, in development 0.4, lets a client inspect repositories and prepare exposure plans without executing project code.