🔌 WebRobot MCP Endpoint

WebRobot exposes an official Model Context Protocol (MCP) server, so any MCP-compatible client or agent (Claude Code, the Claude Agent SDK, Cursor, Claude Desktop, …) can drive the WebRobot ETL platform with tools.

There are two endpoints — pick by what you need:

	🟢 Demo (public)	🔐 Full (authenticated)
Endpoint	https://mcp.webrobot.eu/mcp	https://mcp-full.webrobot.eu/mcp
Transport	Streamable HTTP	Streamable HTTP
Auth	none — public	your own WebRobot API key or platform OAuth token
Scope	demo — only /webrobot/api/demo/*	full — the entire WebRobot REST API (240+ operations)
Surface	stage catalog, generate/validate/execute demo pipelines, designer wizard	everything: datasets, projects, agents, jobs, manifests, RAG, billing, plugins, cloud credentials, …
Data	shared demo sandbox	your org's real data, isolated by RBAC
Best for	trying WebRobot, public demos, no signup	production agents, your own pipelines & data

How the Full endpoint stays safe. It is a pure proxy: it holds no API key of its own. Each request forwards your credential (Authorization: Bearer <token> or X-API-Key) to the WebRobot REST API, which performs all authentication and per-organization RBAC — exactly as if you called the API directly. The MCP never sees more than your key allows.

The Demo server mirrors the public Demo REST API as MCP tools — browse the stage catalog, generate/validate/execute ETL pipelines, run the inference wizard, and manage executions, all from your agent, with no signup.

Connect

Claude Code (CLI)

claude mcp add --transport http webrobot https://mcp.webrobot.eu/mcp

Claude Desktop / Cursor / .mcp.json

{
  "mcpServers": {
    "webrobot": { "type": "http", "url": "https://mcp.webrobot.eu/mcp" }
  }
}

Claude Agent SDK (Python)

from claude_agent_sdk import ClaudeAgentOptions

options = ClaudeAgentOptions(
    mcp_servers={"webrobot": {"type": "http", "url": "https://mcp.webrobot.eu/mcp"}},
    allowed_tools=["mcp__webrobot"],   # allow every WebRobot tool
)

Connect — Full (authenticated)

Use https://mcp-full.webrobot.eu/mcp and send your own credential as a header. The MCP forwards it to the REST API per request (it stores nothing).

.mcp.json (Claude Code / Cursor / Desktop)

{
  "mcpServers": {
    "webrobot-full": {
      "type": "http",
      "url": "https://mcp-full.webrobot.eu/mcp",
      "headers": { "X-API-Key": "<key_id>:<secret>" }
    }
  }
}

Claude Code (CLI)

claude mcp add --transport http webrobot-full https://mcp-full.webrobot.eu/mcp \
  --header "X-API-Key: <key_id>:<secret>"

Claude Agent SDK (Python)

options = ClaudeAgentOptions(
    mcp_servers={"webrobot-full": {
        "type": "http",
        "url": "https://mcp-full.webrobot.eu/mcp",
        "headers": {"X-API-Key": "<key_id>:<secret>"},   # or "Authorization": "Bearer <jwt>"
    }},
    allowed_tools=["mcp__webrobot-full"],
)

Get your API key from the WebRobot portal (Settings → API Keys). Platform OAuth tokens work too — send them as Authorization: Bearer <token>. Your key's role/organization scopes determine exactly which tools succeed.

Tools

The list below is the Demo surface. The Full endpoint auto-exposes the entire WebRobot REST API as tools (datasets, projects, agents, jobs, manifestApply / manifestValidate, RAG query/ingest, cloud credentials, billing, plugins, …) — gated by your key's scopes.

Catalog & pipelinesgetCatalogStages · suggestStages · generatePipeline · saveGeneratedPipeline · generatePythonTransform · validatePythonTransform · securityCheckPythonTransform · uploadDataset · listDemos · reloadPipelines · getPluginInfo

ExecutionexecuteDemo · getExecutionStatus · getExecutionLogs · getExecutionOutput · cancelExecution

Designer wizard (selectors / inference)wizardInferSelector · wizardInferFields · wizardInferSegment · wizardInferActions · wizardSuggestFieldNames · wizardValidate · wizardProxy · cmfOpen · cmfStep · cmfClose

Agentic / BYOCrunDemo · listDemos_1 · executionDetail · validateHetznerToken · cleanupStaleByocSecrets

Demo app servingserveDemoApp · serveStaticFile

Try it

Once connected, ask your agent:

“Using the webrobot tools, list the available ETL stages and summarize five of them.”

Or use the hosted AI Data Engineer chat — it is already wired to this MCP endpoint.

Beta

Both MCP endpoints are evolving — tools and surface may change, and behavior is not yet guaranteed for production use. The Demo endpoint (demo scope) is public and sandboxed; the Full endpoint (full scope) acts on your real data under your own credential, so treat its write tools (datasets, jobs, manifests, billing, plugins) with care.