infranode

# InfraNode

[![License: Apache 2.0](https://img.shields.io/badge/license-Apache_2.0-blue.svg)](./LICENSE)
[![Glama score](https://glama.ai/mcp/servers/street1983nk/infranode/badges/score.svg)](https://glama.ai/mcp/servers/street1983nk/infranode)
[![MCP Registry](https://img.shields.io/badge/MCP_Registry-dev.infranode%2Finfranode-1f6feb)](https://registry.modelcontextprotocol.io)
[![Smithery](https://img.shields.io/badge/Smithery-infranode-7c3aed)](https://smithery.ai/server/infranode/infranode)

**A keyless, remote MCP server for German public-infrastructure open data.**

German cities publish a lot of open data, but every source has its own format,
fields and quirks, and several need portal registration. InfraNode normalizes
~20 categories, weather, air quality, public transit (incl. realtime
departures), traffic, parking, charging, water levels, demographics, energy and
more, for **84+ German cities** behind **one** interface, and serves it as an
MCP server. **38 read-only tools, no API key, no account, no install.**

Sources include the Deutscher Wetterdienst (DWD), Umweltbundesamt (UBA),
Mobilithek/DELFI, GovData, OpenStreetMap, Bundesnetzagentur, KBA, DIVI and more.

## Connect

One line with Claude Code:

```bash
claude mcp add --transport http infranode https://mcp.infranode.dev/mcp
```

Any other MCP client, point it at the remote endpoint (Streamable HTTP):

```jsonc
{
  "mcpServers": {
    "infranode": { "url": "https://mcp.infranode.dev/mcp" }
  }
}
```

- **Cursor / Windsurf:** add the block above to `~/.cursor/mcp.json` (or the app's MCP settings).
- **VS Code:** `code --add-mcp '{"name":"infranode","url":"https://mcp.infranode.dev/mcp"}'`
- **Claude Desktop:** add the same `mcpServers` block to your `claude_desktop_config.json`.
- **ChatGPT:** add a connector with the URL `https://mcp.infranode.dev/mcp`.

Full install guide, the complete tool manifest with example outputs, the
permission model and an example transcript are in
[docs/mcp-install.md](./docs/mcp-install.md). The registry manifest is
[server.json](./server.json).

## Example

Ask your agent something like *"What's the weather in Cologne right now?"* and it
calls `weather`:

```jsonc
// weather(slug="koeln")
{
  "data": {
    "city_slug": "koeln",
    "observed_at": "2026-06-18T13:00:00Z",
    "source": "dwd",
    "attribution": { "text": "Datenbasis: Deutscher Wetterdienst", "modified": true },
    "payload": { "kind": "weather", "temperature_c": 30.4, "humidity": 43.0, "station_id": "02667" }
  },
  "meta": { "source_status": "ok", "cache_status": "hit", "correlation_id": "..." }
}
```

Every response follows the same `{ data, meta }` envelope: each record carries
its `attribution` (license + source), and `meta.source_status` tells you whether
the upstream source delivered data, so a dead source degrades gracefully instead
of failing the call.

> Tip: call `list_cities` first to discover valid city slugs (e.g. `koeln`,
> `berlin`, `hamburg`), then call any city-scoped tool.

## Tools (38)

| Group | Tools |
|-------|-------|
| **Discovery** | `list_cities`, `sources`, `compare` (one resource across many cities) |
| **Weather & environment** | `weather`, `weather_warnings`, `air_quality`, `air_quality_live`, `pollen_uv`, `water_level`, `flood` |
| **Mobility** | `transit`, `transit_departures`, `station_departures`, `station_arrivals`, `traffic`, `road_events`, `webcams`, `charging`, `sharing`, `fuel_prices` |
| **City & people** | `get_city`, `geo`, `demographics`, `indicators`, `unemployment`, `tourism`, `construction`, `accidents`, `health`, `icu_live`, `holidays`, `election`, `events`, `pois` |
| **Energy & vehicles** | `power_load`, `power_price`, `energy`, `vehicle_registrations` |

All tools are annotated `readOnlyHint: true` / `destructiveHint: false` /
`idempotentHint: true`, so MCP clients can safely auto-allow them.

### Prompts & resources

- **Prompts:** `city_briefing`, `compare_air_quality`, `commute_check`, ready-made flows that chain several tools.
- **Resources:** `infranode://cities` and `infranode://sources`, browse the coverage catalog without a tool call.

## How it behaves

- **Keyless & read-only.** No credentials, no writes, no user accounts.
- **Canonical envelope.** `{ data, meta }` with per-source status and attribution.
- **Graceful degradation.** A failing upstream returns `source_status`, not an error.
- **Safe by design.** SSRF and injection gates validate every request; inputs are checked against fixed allowlists.

See [SECURITY.md](./SECURITY.md) for the security model.

## Self-host (optional)

You don't need to, the hosted endpoint above is the fastest path. But the code
is open. Run the API stack locally with Docker (Compose v2):

```bash
cp .env.example .env          # example config, contains NO real secrets
docker compose -f deploy/docker-compose.yml up
curl http://localhost/api/v1/health   # -> {"status":"ok","version":"1.0.0","redis":true}
```

To run the MCP server itself locally over stdio (against the public API):

```bash
uv sync --group mcp
INFRANODE_MCP_API_BASE=https://infranode.dev/api/v1 uv run python -m infranode.mcp.server
```

All settings use the `INFRANODE_` env prefix (see `.env.example`); each data
source is toggled by its own `INFRANODE_ENABLE_*` flag. Real secrets are never
committed, only `.env.example` is versioned and CI runs a gitleaks scan.

## License: code and data are separate

- **Code:** Apache-2.0 (see [LICENSE](./LICENSE)).
- **Data:** the open data served through InfraNode keeps the licenses of its
  upstream sources (e.g. ODbL for OpenStreetMap, DL-DE-BY for GovData, attribution
  for DWD). These data licenses and attribution are tracked separately in
  `DATA-LICENSES.md`. The Apache-2.0 license covers only the API source code, not
  the passed-through data.

## Contributing

Contributions are welcome. Setup, gate commands and the secret rule are in
[CONTRIBUTING.md](./CONTRIBUTING.md).