Skip to content

Changelog

All notable changes to the cotbias library are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.2] - 2026-10-19

Added

  • Comprehensive Google-style docstrings and type annotations across CotClient, AsyncCotClient, resources, models, exceptions, retry transports, and MCP server.
  • MkDocs documentation suite hosted via GitHub Pages with auto-generated API references.
  • Identity guardrails in pre-commit and GitHub Actions CI.

[0.2.1] - 2026-10-19

Changed

  • Packaged pandas as a core dependency for out-of-the-box DataFrame support (client.timeseries.get(..., as_dataframe=True)).
  • Made mcp an optional dependency (pip install "cotbias[mcp]").

[0.2.0] - 2026-10-19

Added

  • Official release of cotbias SDK and MCP Server.
  • Synchronous and asynchronous API clients (CotClient, AsyncCotClient).
  • Resource endpoints:
  • client.timeseries: 156-week positioning data across Legacy, Disaggregated, TFF, and CIT reports.
  • client.inference: AI market regimes, directional bias, and macro attribution.
  • client.screener: Cross-market positioning scans.
  • client.markets: Contract catalogs and CFTC code resolution.
  • Enterprise reliability features:
  • Built-in RetryTransport with exponential backoff and jitter for transient 5xx and timeout errors.
  • Strict Pydantic models with extra="allow" to protect against upstream schema drift.
  • Model Context Protocol (MCP) Server:
  • cotbias-mcp CLI entrypoint.
  • Tools: get_market_snapshot, get_ai_inference, get_timeseries, scan_markets, list_markets.
  • In-memory TTL caching (5m for inference/screener, 1h for timeseries).
  • Prompts and resources for Claude Desktop and Cursor IDE.