Fit-Wiki Scraper & MCP Server

Python tool for extracting course materials, exam variants, and notes from Fit-Wiki. Includes an MCP server for integration with LLM clients (Claude, AGY, OpenCode).

---

Requirements

• Python 3.12+
• Liberation Sans and Liberation Mono fonts (for correct Czech diacritics in PDFs)

Setup

python3 -m venv venv
./venv/bin/pip install -r requirements.txt

Configuration

Copy .env.example to .env and fill in your cookies:

cp .env.example .env

Alternatively, pass FITWIKI_COOKIES as an environment variable in the MCP client config.

---

CLI Usage

Scrape a course to Markdown

./venv/bin/python scraper.py [course_code]

Convert Markdown to PDF

./venv/bin/python convert_to_pdf.py

Compiles all files in markdown_output/ to pdfs/.

Pipeline (single page)

./venv/bin/python index_page.py bi-osy            # download index
./venv/bin/python scraper.py                      # scrape to Markdown
./venv/bin/python convert_to_pdf.py               # compile PDFs

---

MCP Server

The MCP server exposes Fit-Wiki tools to LLM clients. It runs via stdio or streamable-http.

Client Configuration

Add the following to your client's MCP config file. Cookies are loaded from the .env file (see .env.example).

{
  "mcpServers": {
    "fitwiki": {
      "command": "/path/to/fitwiki/venv/bin/python",
      "args": ["/path/to/fitwiki/mcp_server.py"]
    }
  }
}

Config File Locations

Client	Path
Claude Desktop	~/.config/Claude/claude_desktop_config.json
AGY (CLI)	~/.gemini/antigravity-cli/mcp_config.json
AGY (IDE / 2.0)	~/.gemini/config/mcp_config.json
OpenCode	.opencode.json in the workspace root

Restart the client after saving. For AGY, verify with /mcp in chat.

Transports

• stdio (default) – the client spawns the server as a subprocess. Use for Claude Desktop, AGY, OpenCode.
• streamable-http – for Docker or remote setups. Set MCP_TRANSPORT=streamable-http and point the client to http://host:8000/mcp.

---

Docker

Build & Run

docker compose up -d

The container exposes the MCP server on port 8000 via streamable-http. Mounted volumes persist markdown_output/ and pdfs/ on the host.

GitHub Container Registry

The Docker image is built automatically on push (see .github/workflows/docker-build.yml) and published to ghcr.io/<your-username>/fitwiki-mcp. Tags: latest, master, short SHA.

---

Exposed MCP Tools

• list_courses – List all courses on the FIT platform.
• list_course_sections – List sections (e.g. zkouska, test1) for a course.
• list_section_pages – List pages within specific sections.
• download_page – Scrape a page to Markdown and compile to PDF in one step.
• compile_pdf – Compile a Markdown file to PDF.
• compile_category_pdfs – Compile all Markdown files in a category to PDFs.
• read_saved_file – Read a previously saved Markdown file by path.
• scrape_index – Parse a course index and list discovered links.
• scrape_page – Scrape a single page to Markdown and return its content.
• scrape_course – Download all pages in selected categories for a course.
• download_page – Scrape a page to Markdown, compile to PDF, and return the content.