ted-mcp: EU public tenders for AI agents

Read-only MCP server for TED (Tenders Electronic Daily), the EU's official public-procurement journal. It uses TED's free, keyless Search API v3; no account or API key is needed.

Tools

All tools are marked read-only. Notice text is returned as third-party data, and the server tells the AI to treat it as data, not instructions.

Example prompts once connected:

• "How many EU tenders mentioning NIS2 were published since 2026-01-01, vs the same period last year?"

• "List open Romanian IT tenders (CPV 72000000) published this month."

• "Find active tenders in DE, AT and NL mentioning SBOM or 'Cyber Resilience Act'."

Related MCP server: LexSocket MCP Server

1. First run: check it against the real API

python3 -m venv .venv && . .venv/bin/activate
pip install -e ".[dev]"
pytest -q                      # 20 tests, TED is mocked

# Real API smoke test (no key needed):
curl -s -X POST https://api.ted.europa.eu/v3/notices/search \
  -H 'Content-Type: application/json' \
  -d '{"query":"buyer-country=ROU AND publication-date>=20260901","fields":["publication-number","notice-title"],"limit":2}' | head -c 600

Query syntax follows TED's official Expert Search help: single words unquoted (FT ~ kubernetes), phrases quoted (FT ~ "cyber security"), and lists via IN (buyer-country IN (ROU DEU)). Use check_query to validate anything custom before running it.

2. Claude Desktop (stdio, local)

Settings → Developer → Edit Config, then add:

{
  "mcpServers": {
    "ted-tenders": {
      "command": "/ABSOLUTE/PATH/ted-mcp/.venv/bin/ted-mcp"
    }
  }
}

Windows path example: C:\\Users\\you\\ted-mcp\\.venv\\Scripts\\ted-mcp.exe. Restart Claude Desktop.

3. Claude Code

claude mcp add ted-tenders -- /ABSOLUTE/PATH/ted-mcp/.venv/bin/ted-mcp

4. Remote connector (claude.ai web, mobile)

claude.ai custom connectors need a public HTTPS URL. HTTP mode refuses to start without a token.

export TED_MCP_TOKEN=$(openssl rand -hex 24)     # save this in your password manager
docker build -t ted-mcp .
docker run -d --name ted-mcp --restart unless-stopped \
  -e TED_MCP_TOKEN=$TED_MCP_TOKEN \
  -e TED_MCP_ALLOWED_HOSTS=ted.example.com \
  -p 127.0.0.1:8000:8000 ted-mcp

Put TLS in front, for example with Caddy (automatic Let's Encrypt):

ted.example.com {
    reverse_proxy 127.0.0.1:8000
}

Then in claude.ai → Settings → Connectors → Add custom connector:

https://ted.example.com/mcp?token=YOUR_TOKEN

Health check (no token needed): https://ted.example.com/healthz

Security notes

• The token is compared in constant time; requests without it get 401.

• A token in a URL can end up in proxy logs. Keep access logs private and rotate the token if exposed. MCP clients that can send headers should use Authorization: Bearer <token> instead.

• TED_MCP_ALLOWED_HOSTS enables DNS-rebinding protection. Set it to your domain.

• The container runs as a non-root user and exposes only port 8000 on localhost. The server makes outbound calls to TED only.

• Built-in politeness: at most 2 concurrent requests and 0.4 s between calls, with retries on 429/5xx.

Configuration

Data and licence

TED notice metadata is published by the EU Publications Office for free reuse, including commercial use. Keep "Source: TED (ted.europa.eu)" attribution when you republish results. This project is independent and not affiliated with the EU. Code: MIT.