io.github.jztan/qt4-doc-mcp-server
Offline MCP server for local Qt 4.8, Qt 5, and Qt 6 documentation with full-text search.
Open source Open in the app JSON README (API)
About
Offline MCP server for local Qt 4.8, Qt 5, and Qt 6 documentation with full-text search.
Details
- Kind
- MCP servers
- Topic
- No topic detected
- Publisher
- jztan
- Origin
- official
- Category
- ferramentas
- Transport
- local
- Version
- 0.7.0
- Stars
- 3
- Forks
- 2
- Last push
- 2026-07-20T11:39:37Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-08-29 04:00:18
- Updated
- 2026-08-29 04:00:18
- Origin id
io.github.jztan/qt4-doc-mcp-server
README
# Qt Documentation MCP Server
mcp-name: io.github.jztan/qt4-doc-mcp-server
[](https://pypi.org/project/qt4-doc-mcp-server/)
[](LICENSE)
[](https://pypi.org/project/qt4-doc-mcp-server/)
[](https://github.com/jztan/qt4-doc-mcp-server/issues)
[](https://github.com/jztan/qt4-doc-mcp-server/actions/workflows/pr-tests.yml)
[](https://pepy.tech/project/qt4-doc-mcp-server)
Bring locally installed Qt 4.8, Qt 5, or Qt 6 documentation to your AI coding assistant. Works offline with one selected documentation set at a time.
## [Tool Reference](./docs/TOOL_REFERENCE.md) | [Changelog](./CHANGELOG.md) | [Contributing](./docs/CONTRIBUTING.md) | [Troubleshooting](./docs/TROUBLESHOOTING.md)
## ✨ Features
- 🔌 **Offline-First** - Works entirely with local documentation
- 🔍 **Full-Text Search** - Find what you need across all Qt docs
- ⚡ **Smart Caching** - Fast responses for repeated queries
- 🎯 **Fragment Support** - Extract specific sections when needed
- 🛠️ **MCP Standard** - Compatible with Claude, VS Code, and other MCP clients
## 📦 Prerequisites
- **Python 3.11+** required (not needed when using the Docker image)
- **Qt HTML Documentation** for one supported release (Qt 4.8, Qt 5, or Qt 6)
- The included helper downloads Qt 4.8.4 only.
- Point `QT_DOC_BASE` directly at an existing Qt 5/6 offline documentation root.
- **~500MB disk space** for docs + cache + search index
- **SQLite with FTS5 support** (included in Python 3.11+ by default)
## 🚀 Installation
### From PyPI (Recommended)
```bash
pip install qt4-doc-mcp-server
```
### From Source
```bash
git clone https://github.com/jztan/qt4-doc-mcp-server.git
cd qt4-doc-mcp-server
uv sync --locked
```
### Setup Qt Documentation
```bash
# Automated setup (recommended, requires the cloned repo)
python scripts/prepare_qt48_docs.py --segments 4
# This will:
# - Download Qt 4.8.4 source archive
# - Extract HTML documentation
# - Create .env with sensible defaults
# - Copy GFDL license file
```
The helper script ships with the repo, not the PyPI package. Installed from PyPI only? Download the docs manually instead:
```bash
curl -LO https://download.qt.io/archive/qt/4.8/4.8.4/qt-everywhere-opensource-src-4.8.4.tar.gz
tar -xzf qt-everywhere-opensource-src-4.8.4.tar.gz qt-everywhere-opensource-src-4.8.4/doc/html
mv qt-everywhere-opensource-src-4.8.4/doc/html ./qt4-docs-html
```
Then create a `.env` with `QT_DOC_BASE` pointing at that directory (see Configuration below).
### Quick Start Commands
```bash
# 1. Install
pip install qt4-doc-mcp-server
# 2. Setup Qt docs
python scripts/prepare_qt48_docs.py --segments 4
# 3. Build search index (optional; the server builds it on first start)
qt-doc-build-index
# 4. Start server
qt-doc-mcp
# 5. Verify health
curl -s http://127.0.0.1:8000/health
```
The legacy `qt4-doc-mcp-server` command remains available as an alias for existing client configurations.
### Agent-friendly FTS CLI
After building the index, agents can search and receive materialized absolute Markdown paths:
```bash
qt-doc-cli "accessible applications" --limit 5
```
The command reads the same `.env` settings as the server. Before searching, it automatically builds a missing/outdated FTS index and fully warms an incomplete Markdown cache. It then prints each result's title, absolute `.md` path, and FTS snippet to stdout; preparation messages, errors, and warnings go to stderr. Use `qt-doc-warm-md --force` after changing documentation in place.
## 🐳 Docker
Prebuilt multi-arch images (amd64/arm64) are published to GitHub Container Registry on every release. The container is offline-only: you mount your prepared Qt `doc/html` directory read-only at `/docs`, and all derived state (Markdown cache and search index) lives in a volume at `/data`.
```bash
# Prepare Qt docs on the host first (one-time, writes to ./qt4-docs-html)
python scripts/prepare_qt48_docs.py --segments 4
# Run from GHCR
docker run -d --name qt4-doc-mcp-server -p 8000:8000 \
-v "$PWD/qt4-docs-html:/docs:ro" \
-v qt4-doc-data:/data \
ghcr.io/jztan/qt4-doc-mcp-server:latest
# Verify
curl -s http://127.0.0.1:8000/health
```
Using only the published image, without cloning this repo? Download the Qt 4.8.4 docs directly instead of running the prepare script:
```bash
curl -LO https://download.qt.io/archive/qt/4.8/4.8.4/qt-everywhere-opensource-src-4.8.4.tar.gz
tar -xzf qt-everywhere-opensource-src-4.8.4.tar.gz qt-everywhere-opensource-src-4.8.4/doc/html
mv qt-everywhere-opensource-src-4.8.4/doc/html ./qt4-docs-html
```
Any existing Qt 5 or Qt 6 `doc/html` tree (from the Qt installer or distro documentation packages) works as the mount source too.
First start converts and indexes the documentation into the `/data` volume; subsequent starts reuse it. Depending on the docset size, the first start can take a minute or two before the health endpoint responds.
### Docker Compose
With docs in the default `./qt4-docs-html` location, no configuration is needed:
```bash
docker compose up -d
```
For docs elsewhere (including Qt 5 or Qt 6 `doc/html` trees, which the server also supports), point `QT_DOC_HTML_PATH` at them. If port 8000 is taken on the host, set `HOST_PORT` in the same file (with plain `docker run`, change the left side of `-p` instead):
```bash
cp .env.docker.example .env.docker # set QT_DOC_HTML_PATH
docker compose --env-file .env.docker up -d
```
Or use the convenience script, which checks Docker is running, creates `.env.docker` on first run, builds and starts the service, and waits for the health endpoint:
```bash
./deploy.sh
```
Point your MCP client at `http://127.0.0.1:8000/mcp` (streamable HTTP). Qt documentation is licensed under GFDL 1.3; the container serves your local copy and never redistributes it.
## ⚙️ Configuration
These settings apply to native (pip or source) runs. In Docker they are already set inside the container (`QT_DOC_BASE=/docs`, `QT_DOC_STATE_DIR=/data`, `SERVER_HOST=0.0.0.0`); the `QT_DOC_HTML_PATH` variable in `.env.docker` is not a server setting, just the host path mounted at `/docs`.
Create a `.env` file in the repo root. The helper script writes sensible defaults; adjust as needed:
| Variable | Default | Purpose |
| --- | --- | --- |
| `QT_DOC_BASE` | _required_ | Absolute path to one Qt 4.8, Qt 5, or Qt 6 HTML documentation root. The server detects the active docset. |
| `QT_DOC_STATE_DIR` | `$QT_DOC_BASE/.index` | Optional writable directory for the FTS index and Markdown cache. Use this when the documentation root is read-only. |
| `PREINDEX_DOCS` | `true` | Build search index automatically at startup if not present. |
| `PRECONVERT_MD` | `false` | Warm the Markdown cache automatically at MCP startup. |
| `SERVER_HOST` | `127.0.0.1` | Bind address for the FastMCP server (`0.0.0.0` for containers). |
| `SERVER_PORT` | `8000` | TCP port for streamable HTTP transport. |
| `MCP_LOG_LEVEL` | `WARNING` | Logging verbosity (DEBUG/INFO/WARNING/ERROR). |
| `MD_CACHE_SIZE` | `512` | In-memory CachedDoc LRU capacity (counts pages). |
| `DEFAULT_MAX_MARKDOWN_LENGTH` | `20000` | Default maximum characters returned per request (prevents token limit issues). |
The tools identify documents by their exact root-relative Markdown path, not an online URL. For example, use `qcompleter.md` for a Qt 4 page, `qtdoc/accessible.md` for a Qt 5/6 global page, or `qtcore/qobject.md` for a Qt 5/6 Core page. By default, each docset stores its own index and Markdown cache under `$QT_DOC_BASE/.index/`, so switching `QT_DOC_BASE` reuses its existing derived state. Set `QT_DOC_STATE_DIR` to relocate both to a writable directory. The Markdown cache mirrors the documentation tree: for example, `qtcore/qobject.md` is cached as `.index/md/qtcore/qobject.md` plus `qobject.meta.json` with the default state directory.
## 🔌 MCP Client Setup
By default, 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.
### Stdio transport
Run the server over stdio instead of HTTP with:
```bash
qt-doc-mcp --transport stdio
```
For stdio-only MCP clients, configure that command with `args: ["--transport", "stdio"]`. Startup indexing and Markdown-cache progress are written to stderr, leaving stdout exclusively for MCP protocol messages.
<details>
<summary><strong>Visual Studio Code (Native MCP Support)</strong></summary>
VS Code has built-in MCP support via GitHub Copilot (requires VS Code 1.102+).
**Using CLI (Quickest):**
```bash
code --add-mcp '{"name":"qt-docs","type":"http","url":"http://127.0.0.1:8000/mcp"}'
```
**Using Command Palette:**
1. Open Command Palette (`Cmd/Ctrl+Shift+P`)
2. Run `MCP: Open User Configuration` (for global) or `MCP: Open Workspace Folder Configuration` (for project-specific)
3. Add the configuration:
```json
{
"servers": {
"qt-docs": {
"type": "http",
"url": "http://127.0.0.1:8000/mcp"
}
}
}
```
4. Save the file. VS Code will automatically load the MCP server.
**Manual Configuration:**
Create `.vscode/mcp.json` in your workspace (or `mcp.json` in your user profile directory):
```json
{
"servers": {
"qt-docs": {
"type": "http",
"url": "http://127.0.0.1:8000/mcp"
}
}
}
```
</details>
<details>
<summary><strong>Claude Code</strong></summary>
Add to Claude Code using the CLI command:
```bash
claude mcp add --transport http qt-docs http://127.0.0.1:8000/mcp
```
Or configure manually in your Claude Code settings file (`~/.claude.json`):
```json
{
"mcpServers": {
"qt-docs": {
"type": "http",
"url": "http://127.0.0.1:8000/mcp"
}
}
}
```
</details>
<details>
<summary><strong>Codex CLI</strong></summary>
Add to Codex CLI using the command:
```bash
codex mcp add qt-docs -- npx -y mcp-client-http http://127.0.0.1:8000/mcp
```
Or configure manually in `~/.codex/config.toml`:
```toml
[mcp_servers.qt-docs]
command = "npx"
args = ["-y", "mcp-client-http", "http://127.0.0.1:8000/mcp"]
```
**Note:** Codex CLI primarily supports stdio-based MCP servers. The above uses `mcp-client-http` as a bridge for HTTP transport.
</details>
<details>
<summary><strong>Kiro</strong></summary>
Kiro primarily supports stdio-based MCP servers. For HTTP servers, use an HTTP-to-stdio bridge:
1. Create or edit `.kiro/settings/mcp.json` in your workspace:
```json
{
"mcpServers": {
"qt-docs": {
"command": "npx",
"args": [
"-y",
"mcp-client-http",
"http://127.0.0.1:8000/mcp"
],
"disabled": false
}
}
}
```
2. Save the file and restart Kiro. The active Qt documentation tools will appear in the MCP panel.
**Note:** Direct HTTP transport support in Kiro is limited. The above configuration uses `mcp-client-http` as a bridge to connect to HTTP MCP servers.
</details>
<details>
<summary><strong>Generic MCP Clients</strong></summary>
Most MCP clients use a standard configuration format. For HTTP servers:
```json
{
"mcpServers": {
"qt-docs": {
"type": "http",
"url": "http://127.0.0.1:8000/mcp"
}
}
}
```
For clients that require a command-based approach with HTTP bridge:
```json
{
"mcpServers": {
"qt-docs": {
"command": "npx",
"args": ["-y", "mcp-client-http", "http://127.0.0.1:8000/mcp"]
}
}
}
```
</details>
## 🛠️ Available Tools
The server provides **2 MCP tools** for working with the active local Qt documentation set:
1. **`read_documentation`** - Read and convert pages from the active Qt documentation set to Markdown
- Fragment extraction (`#details`, `#public-functions`)
- Pagination with `start_index` and `max_length`
- Section-only mode for targeted content
- Returns Markdown with normalized links and GFDL attribution
2. **`search_documentation`** - Full-text search across the active Qt documentation set
- SQLite FTS5 with BM25 relevance ranking
- Context snippets with highlighted matches
- Configurable result limits (default: 10, max: 50)
For detailed API documentation including parameters, return values, examples, and error handling, see the **[Tool Reference](docs/TOOL_REFERENCE.md)**.
## 📚 Related Resources
- [Model Context Protocol Specification](https://modelcontextprotocol.io/)
- [Qt Documentation](https://doc.qt.io/)
- [FastMCP Framework](https://github.com/jlowin/fastmcp)
- [Tool Reference](docs/TOOL_REFERENCE.md)
- [Changelog](CHANGELOG.md)
- [Contributing Guide](docs/CONTRIBUTING.md)
- [Troubleshooting Guide](docs/TROUBLESHOOTING.md)
## 📄 License
- **Code:** MIT License (see `LICENSE`).
- **Qt Documentation:** © The Qt Company Ltd. and contributors, 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 details.