# Documentation MCP Server Rules

## Context
This project uses the `docs-mcp-server` to provide grounded, real-time access to technical documentation. These rules ensure the AI prioritizes this live data over internal training knowledge.

## Rules for Documentation Retrieval
1. **Mandatory Documentation Check**: Before answering any question regarding API signatures, library configurations, or framework versions, you MUST use the `docs-mcp-server` tools (e.g., `search_docs` or `ask_question`).
2. **Preference for Retrived Content**: If the information retrieved from the MCP server contradicts your internal training data, the MCP data is the "source of truth." You must follow it strictly.
3. **No Hallucinations**: If the MCP server returns no results for a specific query, do not guess. State "No official documentation found via MCP" and ask the user if they would like you to proceed based on general knowledge or if they have a local file to reference.
4. **Citation Requirement**: When providing an answer based on MCP data, include a brief mention that the information was retrieved from the live documentation server.

## Operational Workflow
- **Search First**: Start by searching for broad keywords related to the user's request.
- **Drill Down**: If multiple results are found, use specific queries to narrow down the relevant documentation sections.
- **Verify Config**: Use the server to check for `opencode.json` schema requirements or tool-specific parameters before suggesting configuration changes.