OuterSpace Apizr

What OuterSpace Apizr does

OuterSpace Apizr MCP server works with existing Python code, including training scripts, notebooks, and independent functions. It provides a capability-compilation workflow: inspect code, identify usable functions, select the functions that should be public, generate their interfaces, and define how they execute. The original business logic remains in Python.

The project supports two related paths. An optional experiment workflow records local runs and their evidence, including inputs such as data, parameters, environment details, metrics, and outputs. The exposure workflow can be used independently, so a function does not need a completed experiment run before it is made available through MCP or REST.

How it works

The process begins with source inspection. The apizr experiment inspect command can examine a script without executing it and report code structure, data, parameters, randomness, environment information, metrics, outputs, and possible serving candidates. Apizr 0.4.5 also recognizes supported Python TypedDict payload models, including nested JSON inputs for generated REST and MCP interfaces.

When trusted execution is appropriate, Apizr can run local workloads and retain run history. Saved runs can be listed, inspected individually, and compared. Comparisons identify recorded changes and calculate numeric deltas, but they do not establish which change caused a metric difference.

The MCP analysis server supports a compact summary view and a detailed view for an individual capability. Detail output can include dependency evidence, pagination, and explanations when a requested exposure is refused.

Setup and configuration

Install Apizr 0.4.5 with uv in a Python 3.11–3.14 environment:

uv tool install outerspace-apizr==0.4.5

The core package and optional MCP, OCI, and Attest plugins use the same version. Optional plugins run in separate environments, and installing or activating them is independent from operator authorization. Consult the project’s installation and release documentation for version-specific functionality and verified publication details.

Tools and capabilities

OuterSpace Apizr MCP server provides capabilities for:

• Inspecting Python experiments without running them.
• Recording trusted local runs and retaining local history.
• Listing and showing recorded runs.
• Comparing evidence from two validated local runs.
• Analyzing project capabilities through summary and detail views.
• Selecting Python functions for MCP or REST exposure.
• Handling supported structured TypedDict JSON inputs.

Limitations and notes

Local execution is not a security sandbox. Code run through the experiment workflow has access to the host filesystem, network, and subprocesses, so only trusted workloads should be executed.

Static analysis and generation cover supported Python patterns; generated servers have their own runtime dependencies. TypedDict model support has declaration limits described in the project documentation. Exposure refusals can be localized, and unfinished or unrelated experiments do not necessarily prevent an independent function from being exposed.

The comparison feature reports differences in recorded evidence rather than causal explanations. It also relies on runs from the same validated local store. The provided material does not document specific MCP client compatibility or client-specific setup instructions.