ocds-mcp

MCP server for German public procurement data (OCDS). Connects your AI assistant (Claude, GPT, etc.) to the Vergabe Dashboard API for semantic search, tender matching, and company profile management.

Your company profiles never leave your machine — only embedding vectors are sent to the API. GDPR-compliant by design.

Quick Start

1. Get an API key

Sign up at vergabe-dashboard.qune.de and create an API key (MCP or Enterprise plan required).

2. Install

Via npx (easiest — downloads the correct binary automatically):

npx @qune-tech/ocds-mcp --api-key sk_live_YOUR_KEY_HERE

Or download pre-built binary from GitHub Releases:

Platform	Download
Linux x86_64	ocds-mcp-linux-x86_64.tar.gz
macOS Apple Silicon	ocds-mcp-macos-arm64.tar.gz
Windows x86_64	ocds-mcp-windows-x86_64.zip

Linux / macOS:

# Example for Linux x86_64 — adjust the filename for your platform
tar xzf ocds-mcp-linux-x86_64.tar.gz
sudo mv ocds-mcp-linux-x86_64 /usr/local/bin/ocds-mcp

Windows: Extract the zip and move ocds-mcp-windows-x86_64.exe somewhere on your PATH (e.g. C:\Users\YOU\.local\bin\ocds-mcp.exe).

Or build from source:

git clone https://github.com/qune-tech/ocds-mcp.git
cd ocds-mcp
cargo build --release
# Binary at target/release/ocds-mcp

3. Configure your AI client

Claude Desktop — edit claude_desktop_config.json:

Using npx:

{
  "mcpServers": {
    "ocds": {
      "command": "npx",
      "args": ["-y", "@qune-tech/ocds-mcp", "--api-key", "sk_live_YOUR_KEY_HERE"]
    }
  }
}

Using the binary directly:

{
  "mcpServers": {
    "ocds": {
      "command": "ocds-mcp",
      "args": ["--api-key", "sk_live_YOUR_KEY_HERE"]
    }
  }
}

Claude Code — add .mcp.json to your project root:

Using npx:

{
  "mcpServers": {
    "ocds": {
      "command": "npx",
      "args": ["-y", "@qune-tech/ocds-mcp", "--api-key", "sk_live_YOUR_KEY_HERE"]
    }
  }
}

Using the binary directly:

{
  "mcpServers": {
    "ocds": {
      "command": "ocds-mcp",
      "args": ["--api-key", "sk_live_YOUR_KEY_HERE"]
    }
  }
}

Cursor — Settings → MCP Servers → Add:

Using npx:

• Command: npx
• Args: -y @qune-tech/ocds-mcp --api-key sk_live_YOUR_KEY_HERE

Using the binary directly:

• Command: ocds-mcp
• Args: --api-key sk_live_YOUR_KEY_HERE

LM Studio — Settings → MCP → Add Server:

1. Click + Add Server and choose STDIO
2. Fill in:

Using npx:

• Name: ocds
• Command: npx
• Arguments: -y @qune-tech/ocds-mcp --api-key sk_live_YOUR_KEY_HERE

Using the binary directly:

• Name: ocds
• Command: full path to the binary, e.g. /usr/local/bin/ocds-mcp
• Arguments: --api-key sk_live_YOUR_KEY_HERE

3. Click Save
4. In the chat, select a model that supports tool use and enable the ocds server

LM Studio requires models with tool-calling support (e.g. Qwen 2.5, Mistral, Llama 3.1+). Smaller models may not use all 10 tools reliably — 7B+ recommended.

Replace sk_live_YOUR_KEY_HERE with your actual API key.

Available Tools

Tool	Description
search_text	Semantic search across all tenders
list_releases	Filter and browse tenders by month, CPV code, category, value range
get_release	Full tender details by OCID
get_index_info	Database statistics and connectivity check
create_company_profile	Create a matching profile for your company
update_company_profile	Update an existing profile
get_company_profile	View profile details
list_company_profiles	List all your profiles
delete_company_profile	Delete a profile
match_tenders	Match a profile against all tenders with semantic similarity

CLI Options

Usage: ocds-mcp [OPTIONS]

Options:
      --db <DB>            Local profiles database [default: profiles.db]
      --data-dir <DIR>     Data directory [default: data]
      --api-url <URL>      Vergabe Dashboard API [default: https://vergabe-dashboard.qune.de]
      --api-key <KEY>      API key [env: OCDS_API_KEY]
  -h, --help               Print help

How It Works

LLM ←stdio→ ocds-mcp (local)
               │  Local: company profiles + sentence embedder
               │  Remote: searches, release queries
               └──HTTPS──→ Vergabe Dashboard API

The MCP server runs locally on your machine:

• Company profiles are stored in a local SQLite database — they never leave your network.
• Text embeddings are computed locally using a multilingual ONNX model (multilingual-e5-small, ~118 MB, auto-downloaded on first use).
• Only embedding vectors (arrays of 384 floats) are sent to the API for search and matching — your profile text stays local.
• Tender data is fetched from the API on demand.

Requirements

• An API key from vergabe-dashboard.qune.de (MCP or Enterprise plan)
• ~200 MB disk space for the ONNX model (downloaded automatically on first run)
• Internet connection to reach the API

License

MIT — see LICENSE.