Block 8 of the Traceo production readiness epic adds three AI-powered capabilities to the MCP server, each exposed as a tool that any AI coding assistant can call:Semantic Search — Ask “find requirements about authentication” and get results ranked by meaning, not keywords. Every requirement gets a vector embedding on create or update. When a user searches, the query is embedded and compared via cosine similarity against all requirements in their workspace. If no embeddings exist yet, it falls back gracefully to keyword search.Quality Analysis — Pass a requirement (or raw text) to the analyser and get a structured quality report: completeness score, ambiguity score, testability score, consistency score, specific issues found, and actionable suggestions. Batch analysis across multiple requirements produces aggregate statistics with quality distribution. The LLM returns structured JSON — no free-text parsing, no prompt fragility.RAG Document Q&A — Ingest project documentation into chunks, embed each chunk, and then ask natural language questions. The system retrieves the most relevant chunks, passes them as context to an LLM, and returns an answer with source citations. If there’s not enough context, it says so instead of hallucinating.
The entire block — 7 GitHub issues, 3 services, 3 MCP tools, 1 database migration, 32 tests — was implemented in a single FORGE-driven session. The N8N workflow identified Block 8 as the next unblocked block, assembled the implementation prompt with all task details and acceptance criteria, and handed it to Claude Code.Plan mode exploration launched parallel agents to understand the existing server architecture. A plan was produced, approved, and executed across 5 tasksets. Every test passes in CI without an API key — the mock client uses deterministic hash-based embeddings so tests verify real logic without touching OpenAI.
Requirements management tools have a fundamental problem: the more requirements you have, the harder it is to find the one you need. Keyword search fails when people use different words for the same concept. Semantic search solves this by matching on meaning.Quality analysis addresses a different gap: requirements rot. They start precise and gradually accumulate vague language, missing criteria, and internal contradictions. Manual review doesn’t scale. Automated analysis that flags “this requirement has no measurable acceptance criteria” catches drift early.RAG closes the loop between documentation and decision-making. Instead of hunting through a wiki for “what’s our policy on data retention,” you ask the question and get an answer with citations pointing to the exact source paragraphs.
Block 8 lays the AI foundation. Blocks 9-16 build on it: versioning and change history, import/export, the SRS traceability engine, observability, and the frontend experience layer. The semantic search and RAG infrastructure from Block 8 will underpin the knowledge command centre in Block 21.Traceo.cat. Requirements that understand themselves.
