Scripting Docs

You can run a small MCP server that exposes a dedicated context for ScriptingApp documentation. This server ingests multilingual Markdown content, builds a persistent vector store, and serves top-k context through a standard MCP interface so you can query it from your terminal-based LLMs or client CLIs.

How to use

Start the MCP server to expose a query endpoint named scripting_docs. Use your MCP client to connect and invoke the server’s query tool. The server reads from a shared persistent index and returns the top-k relevant document chunks along with their metadata. You can feed those results into terminal LLMs or pipe them into CLI models.

How to install

Prerequisites you need on your machine: Python 3.10+ (validated on 3.12/3.13), a functional MCP environment (uv/uvx), and access to the Markdown documentation assets you intend to ingest. If you want to use OpenAI embeddings, prepare OPENAI_API_KEY.

Install or prepare the tooling and dependencies for MCP server operation. Choose one of the following options to prepare your environment and run the server.

uv run --quiet scripts/mcp_docs_server.py --persist-dir storage/llamaindex
# or to run via a remote MCP client:
uvx --from git+https://github.com/JaxsonWang/Scripting-Docs-MCP mcp-docs-server \
  --persist-dir storage/llamaindex --quiet

Configuration and running notes

The MCP server uses a persistent index stored under storage/llamaindex. The default chunking size and embedding backend can be tuned during ingestion, but the server expects to load the same index you created during ingestion.

To customize how queries are resolved, you can adjust the default top-k returned by the server with the default-k setting in the server command. The server exposes a single endpoint named scripting_docs_query for cllient integrations.

Troubleshooting and tips

If you need to switch embedding backends or models, re-ingest the documentation to rebuild the index with the new backend or model settings.

Ensure your OPENAI_API_KEY (if using OpenAI embeddings) is set in your environment so the embedding step can access the API.

MCP endpoint configuration (example)

{
  "servers": {
    "scripting_docs": {
      "command": "uv",
      "args": [
        "run",
        "--quiet",
        "scripts/mcp_docs_server.py",
        "--persist-dir",
        "storage/llamaindex"
      ]
    }
  }
}

Tools and endpoints

You gain access to a dedicated MCP endpoint named scripting_docs_query. This tool returns the top-k document chunks and their metadata in response to a user question.

Available tools

scripting_docs_query

MCP endpoint that retrieves top-k document chunks and formats contextual information for client LLMs or CLI models.

ingest_docs

Ingest and chunk multilingual Markdown content into a persistent vector store for querying.

query_docs

Query the built index and format results for CLI models or MCP clients.

mcp_docs_server

MCP server that exposes the scripting_docs_query tool via stdio for integration with Codex/Claude/Gemini CLI.