Skip to content

Releases: SuperagenticAI/supercodemode

🐍v0.1.0: First Public Release (GEPA-centric Code Mode Optimization for MCP)

22 Feb 16:50

Choose a tag to compare

SuperCodeMode v0.1.0

First public release of SuperCodeMode: a Python CLI and demo harness for optimizing Code Mode-style
client behavior in MCP workflows with GEPA
.

SuperCodeMode focuses on improving the client-side text and routing policy around a small tool surface
(typically discovery + execution), so agents make better tool choices and produce more reliable results
without requiring server-side changes.

Highlights

GEPA-centric Code Mode optimization

  • Optimize Code Mode client behavior with GEPA
  • Improve:
    • system prompt text
    • Code Mode description / routing guidance
    • tool alias mappings
    • tool description overrides

MCP runtimes and transports

  • MCP stdio runner for local workflows
  • MCP streamable HTTP runner for direct Cloudflare MCP
  • HTTP bridge runner for custom runtimes

Execution backends (demo MCP server)

  • local
  • docker
  • monty (Python-native sandbox path)

Strategy benchmarking

  • New scm benchmark command compares:
    • tool_call (naive execution-first policy)
    • codemode_baseline
    • codemode_optimized

Observability (shared event schema)

Backends:

  • jsonl
  • otlp
  • logfire (optional)
  • mlflow (optional)
  • langsmith (optional)
  • langfuse (optional)

Includes:

  • CLI context fields (cli_command, cli_runner, cli_executor_backend)
  • summary events (showcase, benchmark, optimize)
  • summary artifacts for CI-friendly comparisons

Docs and examples

  • MkDocs Material documentation site
  • runnable examples for:
    • Cloudflare MCP
    • local MCP stdio
    • optimization flows
    • Gemini demo

Install

pip install supercodemode

With uv (CLI install):

uv tool install supercodemode

Optional extras:

pip install "supercodemode[monty]"
pip install "supercodemode[observability]"

## Quick Start

scm doctor
scm showcase --runner mcp-http
scm showcase --runner mcp-stdio
scm benchmark --runner mcp-stdio

## Links

- Docs: https://superagenticai.github.io/supercodemode/
- PyPI: https://pypi.org/project/supercodemode/
- GEPA docs: https://gepa-ai.github.io/gepa/
- Cloudflare Code Mode MCP blog: https://blog.cloudflare.com/code-mode-mcp/
- UTCP Code Mode implementation: https://github.com/universal-tool-calling-protocol/code-mode

## Notes

- Observability provider SDK backends are best-effort and optional. Telemetry failures do not fail the main
  run path.
- This repo is the end-to-end experimentation/demo harness for the GEPA Code Mode adapter work.