docs-mcp-server

# Spec: Search Quality Evaluation ## ADDED Requirements ### Requirement: Automated Search Evaluation The system MUST provide an automated pipeline to evaluate search result quality against a known dataset. #### Scenario: Running the evaluation suite - **Given** the `docs-mcp-server` has indexed the "react" documentation - **When** the developer runs `npm run evaluate:search` - **Then** the system should execute the `promptfoo` evaluation suite - **And** it should generate a report comparing search results against expected golden answers. #### Scenario: Assessing Relevance - **Given** a search result for "How to use useEffect" - **When** the result is evaluated - **Then** the LLM judge should score it high if it mentions "side effects" and "dependency array" - **And** score it low if it discusses `useState` or class components. #### Scenario: Assessing Contextual Integrity - **Given** a search result with a code block - **When** the integrity check runs - **Then** it must fail if the code block start tag (` ``` `) is present but the end tag is missing. #### Scenario: Assessing Ranking - **Given** a set of 5 search results - **When** the ranking check runs - **Then** it must verify that the `expectedUrl` from the dataset is present in the top 3 results. #### Scenario: Handling missing index - **Given** the "react" documentation is NOT indexed - **When** the evaluation starts - **Then** it should fail fast with a clear error message instructing the user to run `docs-mcp-server scrape ...` first.