Skip to content

Context7 Semantic Indexer & Local Snippet Search

Purpose

This skill governs the integration of Context7 within the DSOM architecture. Context7 acts as both an external semantic retrieval engine (RAG) and a local code snippet database for the Sovereign Markdown Palace (docs/). It parses markdown documentation, chunks it, and generates searchable code/command snippets using LLM extraction.

Integration Guidelines

1. The Indexing Process

Based on our project structure, Context7 targets the docs/ folder: - Context7 clones the main branch. - It bypasses non-RST and non-code configurations to focus directly on Open Knowledge Format (OKF) markdown documentation.

2. Utilizing Context7 via MCP

For AI agents (Cursor, Google Jules, Claude) to query this semantic index: 1. Ensure the agent's MCP (Model Context Protocol) is configured to connect to https://mcp.context7.com/mcp. 2. The agent must inject the CONTEXT7_API_KEY directly into the MCP client configuration via environment variables.

3. Triggering Re-indexing

Context7 automatically syncs with the configured repository branch. If a massive structural change occurs in the Palace, ensure the changes are fully committed and pushed to main so Context7's webhook can detect the diff.

4. Direct LLM Endpoint & MCP Stream Access

For external AI tools (Cursor, Claude, ChatGPT, Gemini, Google Jules), Context7 exposes the live, compiled token stream: - Canonical Stream Endpoint: https://context7.com/gitlab_linuxmalaysia/deep-state-of-mind-for-my-ai/llms.txt?tokens=83688 - MCP Native Tool: FastMCP exposes the fetch_context7_stream tool inside tools/mcp/server.py to stream or query Context7 snippets dynamically via stdio. - Token Adjustments: Adjust the ?tokens=... parameter (e.g. 10000, 30000, 83688) based on the target LLM's context window budget.

5. Local Code Snippet Search (search_code_snippets)

To avoid token bloat and query executable terminal commands quickly without external network round-trips: - FastMCP exposes the search_code_snippets(query, limit) tool in tools/mcp/server.py. - It searches the parsed blocks in references/llms-from-context7.txt and returns the most relevant code blocks and instructions. - A human-readable compiled cheatsheet is maintained in docs/reference/CLI-QUICK-REFERENCE.md.

6. Offline Fallback & Local Snapshot Maintenance

When operating offline or in air-gapped environments: - The local snapshot references/llms-from-context7.txt serves as the primary fallback. - Calling fetch_context7_stream(return_offline_sample=True) provides immediate local context preview without internet access.

7. Automated & Manual Snapshot Updating

  • Manual Invalidation & Fetch: Run the dedicated tool with desired token budget: 
    python tools/fetch_context7_snapshot.py --tokens 250000
  • Automated Scheduled Sync: .github/workflows/context7-sync.yml automatically executes every Sunday at 02:00 UTC (or on-demand via workflow_dispatch), pulling the latest 250k token snapshot using the provisioned CONTEXT7_API_KEY secret.

Security (Rule 24 Mandate)

  • NEVER write the CONTEXT7_API_KEY (e.g., ctx7sk-...) to a local .env file, .git/config, or any file tracked by Git.
  • GitLab CI/CD: If Context7 requires integration via GitLab CI, the API key MUST be stored in GitLab CI/CD Settings as a masked and protected variable.

Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-08-23 Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0