v0.1.0 — First Stable Baseline

What's in this release

First end-to-end tested baseline of Context Blocks — a toolkit for building domain knowledge bases that serve AI agents.

Full Pipeline

cb init → cb extract → cb dedup → cb eval → cb ask → cb viewer → cb mcp

Highlights

• Multi-provider LLM support — Anthropic, OpenAI, Gemini via litellm. Set LLM_PROVIDER and LLM_MODEL env vars.
• MCP server works without API key — retrieval-only mode returns ranked entities + context for the calling agent (Copilot, Claude Desktop, Cursor) to synthesize. No double LLM call.
• Async MCP — compatible with both stdio and streamable-http transports.
• Domain-Aware Retrieval (DAR) — 7-stage pipeline: intent classification, vector + keyword + graph search, RRF fusion, context assembly, synthesis, gap detection.
• Interactive viewer — Astro-based UI for exploring entities, relationships, and coverage gaps.
• Eval framework — persona-based question generation + automated coverage scoring (CLEAN / INCOMPLETE / MISSING).

Install

pip install git+https://github.com/ea-toolkit/context-blocks.git@v0.1.0

Stats

• 171 tests passing
• 6 MCP tools: list_blocks, get_overview, search_entities, get_entity, ask_kb, get_gap_report
• Tested with Anthropic Sonnet + Haiku, OpenAI GPT-4o, Google Gemini