Qt4 Documentation Server

PyPI Version License Python Version GitHub Issues CI

Offline‑only MCP Server that serves Qt 4.8.4 documentation to Agents/LLMs and IDEs. It loads local HTML docs, converts pages to Markdown, and provides fast full‑text search via SQLite FTS5.

Quickstart

1. Install the package: pip install qt4-doc-mcp-server.
2. Fetch and stage the Qt docs (one-time): python scripts/prepare_qt48_docs.py --segments 4.
3. Copy .env from the script output or create one manually (see table below).
4. Build the search index: qt4-doc-build-index (required for search functionality).
5. Run the server: qt4-doc-mcp-server (or uv run python -m qt4_doc_mcp_server.main).
6. Verify health: curl -s http://127.0.0.1:8000/health → { "status": "ok" }.
7. Optional: warm the Markdown cache for faster responses: qt4-doc-warm-md.

Project Structure

.
├─ README.md                    # Quick start, config, licensing
├─ LICENSE                      # MIT license for this codebase
├─ CHANGELOG.md                 # Keep a Changelog (Unreleased + releases)
├─ THIRD_PARTY_NOTICES.md       # Qt docs and deps licensing notes
├─ pyproject.toml               # Packaging, deps, console entry points
├─ scripts/
│  ├─ prepare_qt48_docs.py      # Download, extract, and stage Qt 4.8.4 docs; writes .env
├─ src/
│  └─ qt4_doc_mcp_server/
│     ├─ __init__.py            # Package version
│     ├─ main.py                # FastMCP app (+ /health) and CLI run()
│     ├─ config.py              # Env loader (dotenv) + startup checks
│     ├─ tools.py               # MCP tools (read_documentation, search_documentation)
│     ├─ fetcher.py             # Canonical URL + local path mapping
│     ├─ convert.py             # HTML extraction, link normalization, HTML→Markdown
│     ├─ cache.py               # LRU + Markdown store (disk) helpers
│     ├─ doc_service.py         # Read path orchestration (store + convert)
│     ├─ search.py              # FTS5 index build/query with BM25 ranking
│     └─ cli.py                 # CLI utilities (qt4-doc-warm-md, qt4-doc-build-index)
└─ tests/                       # pytest suite (e.g., test_doc_service.py)

Requirements

• Python 3.11+
• Local Qt 4.8.4 HTML documentation (see below)

Get the Qt 4.8.4 Docs

Prepare Docs with Python helper (recommended)

python scripts/prepare_qt48_docs.py # copy docs by default into ./qt4-docs-html

OR

python scripts/prepare_qt48_docs.py --segments 4 # faster download with 4 segments

This will:

• Download and extract the Qt 4.8.4 source archive (or reuse if present)
• Stage the HTML docs at qt4-docs-html (symlink by default)
• Copy LICENSE.FDL next to the docs
• Create/update .env with QT_DOC_BASE and sensible defaults

Configure (dotenv)

Create a .env file in the repo root. The helper script writes sensible defaults; adjust as needed:

Dev Setup and Run

uv venv .venv && source .venv/bin/activate

# Option 1: run without installing the package (dev-only)
# Using uv to run the module directly
uv run python -m qt4_doc_mcp_server.main

# Option 2: install and use the CLI
uv pip install -e .[dev]
qt4-doc-mcp-server
# Health check
curl -s http://127.0.0.1:8000/health

# Build the search index (required for search_documentation tool)
uv run qt4-doc-build-index

# Optional: preconvert all HTML→Markdown into the store for faster reads
uv run qt4-doc-warm-md

# Run tests (ensure TMPDIR points to a writable location when sandboxed)
uv run python -m pytest -q

How It Works (high‑level)

• Offline‑only: no external HTTP fetches; everything reads from QT_DOC_BASE.
• HTML→Markdown: focused extraction of main content; normalized internal links; attribution appended.
• Markdown store: preconverted pages saved under .cache/md (sharded by URL hash) for fast reads; in‑memory LRU caches hot pages.
• Search: SQLite FTS5 index (title/headings/body) with BM25 ranking and context snippets.

MCP Tools

The server provides two MCP tools:

1. read_documentation - Read and convert a specific Qt documentation page
2. search_documentation - Search across all Qt 4.8.4 documentation

Example: read_documentation

Example MCP request/response (trimmed for brevity):

// request
{
  "method": "tools/run",
  "params": {
    "name": "read_documentation",
    "arguments": {
      "url": "https://doc.qt.io/archives/qt-4.8/qstring.html",
      "fragment": "#details",
      "section_only": true,
      "max_length": 2000
    }
  }
}

// response
{
  "result": {
    "title": "QString Class",
    "canonical_url": "https://doc.qt.io/archives/qt-4.8/qstring.html",
    "markdown": "# QString Class\n...",
    "links": [
      {"text": "QStringList", "url": "https://doc.qt.io/archives/qt-4.8/qstringlist.html"}
    ],
    "attribution": "Content © The Qt Company Ltd./Digia — GNU Free Documentation License 1.3",
    "content_info": {
      "total_length": 15234,
      "returned_length": 2000,
      "start_index": 0,
      "truncated": true
    }
  }
}

Note: The content_info field appears when content is paginated or truncated. Use start_index and max_length parameters to retrieve additional pages. By default, responses are limited to 20,000 characters to avoid exceeding LLM token limits.

Example: search_documentation

Example MCP request/response for searching:

// request
{
  "method": "tools/run",
  "params": {
    "name": "search_documentation",
    "arguments": {
      "query": "signals slots",
      "limit": 5
    }
  }
}

// response
{
  "result": {
    "query": "signals slots",
    "count": 5,
    "results": [
      {
        "title": "Signals and Slots",
        "url": "https://doc.qt.io/archives/qt-4.8/signalsandslots.html",
        "score": 12.34,
        "context": "…used for communication between objects. <b>Signals</b> and <b>slots</b> mechanism is a central…"
      },
      {
        "title": "QObject Class Reference",
        "url": "https://doc.qt.io/archives/qt-4.8/qobject.html",
        "score": 8.76,
        "context": "…The QObject class supports <b>signals</b> and <b>slots</b> for inter-object communication…"
      }
    ]
  }
}

Notes:

• Search uses SQLite FTS5 with BM25 ranking for relevance
• Context snippets highlight matching terms with <b> tags
• The limit parameter controls maximum results (default: 10, max: 50)
• Build the index first with qt4-doc-build-index or set PREINDEX_DOCS=true

MCP Client Configuration

The server exposes an HTTP endpoint at http://127.0.0.1:8000/mcp. Register it with your preferred MCP-compatible agent using the instructions below.

code --add-mcp '{"name":"qt4-docs","type":"http","url":"http://127.0.0.1:8000/mcp"}'

{
  "servers": {
    "qt4-docs": {
      "type": "http",
      "url": "http://127.0.0.1:8000/mcp"
    }
  }
}

claude mcp add --transport http qt4-docs http://127.0.0.1:8000/mcp

{
  "mcpServers": {
    "qt4-docs": {
      "type": "http",
      "url": "http://127.0.0.1:8000/mcp"
    }
  }
}

codex mcp add qt4-docs -- npx -y mcp-client-http http://127.0.0.1:8000/mcp

[mcp_servers.qt4-docs]
command = "npx"
args = ["-y", "mcp-client-http", "http://127.0.0.1:8000/mcp"]

{
  "mcpServers": {
    "qt4-docs": {
      "type": "http",
      "url": "http://127.0.0.1:8000/mcp"
    }
  }
}

{
  "mcpServers": {
    "qt4-docs": {
      "command": "npx",
      "args": ["-y", "mcp-client-http", "http://127.0.0.1:8000/mcp"]
    }
  }
}

Deployment

• Direct (systemd, bare metal, CI runners):
  • Install with pip install qt4-doc-mcp-server.
  • Ensure .env points to your Qt docs and writable cache/index directories.
  • Start with qt4-doc-mcp-server; add PRECONVERT_MD=true for faster first reads.
• Containerization (roadmap):
  • Docker support is planned; follow the repository for updates or open an issue if you need it sooner.

Licensing

• Code: MIT License (see LICENSE).
• Qt docs: © The Qt Company Ltd./Digia, licensed under GFDL 1.3. This server converts locally obtained docs and includes attribution in outputs. If you redistribute a local mirror, include LICENSE.FDL and preserve notices.
• See THIRD_PARTY_NOTICES.md for more.