claude-code-webrobot-skills
WebRobot-Ltd/claude-code-webrobot-skills
Instalação
npx convez-ai WebRobot-Ltd/claude-code-webrobot-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
🤖 WebRobot — Claude Code plugin & MCP server
Turn Claude Code into a data engineer — scrape the web, build Spark ETL pipelines, and run them, all by conversation.
🎬 Demo GIF coming soon — meanwhile try the ▶ live demo.
WebRobot is "ETL for AI Agents" — a Spark-native, EU-sovereign (BYOC) data platform that AI agents operate end to end: from a URL to an insight in a conversation. This plugin is the front door from Claude Code.
It bundles:
- 10 skills — invokable as slash commands inside Claude Code — covering pipelines, the CLI, the public SDKs, plugin development, and the platform overview.
- A hosted Model Context Protocol (MCP) server that exposes the WebRobot REST API as typed tools Claude can call directly (list / create / execute / inspect projects, jobs, pipelines, agents, datasets, cloud credentials, LLM providers, plus manifest validate/apply and ETL stage discovery).
Repo: github.com/WebRobot-Ltd/claude-code-webrobot-skills
— ships as .claude-plugin/plugin.json + skills/ + mcp-server/.
⭐ If this is useful, star the repo — it helps other data & AI engineers find it.
Install
In Claude Code, add this repository as a plugin source and install the
webrobot plugin:
/plugin marketplace add https://github.com/WebRobot-Ltd/webrobot-claude-plugin
/plugin install webrobot
Claude Code will pick up the skills (under /) and connect the webrobot
MCP server (HTTP) from the mcpServers.webrobot entry in plugin.json —
the hosted, official endpoint https://mcp.webrobot.eu/mcp. Nothing to
install locally (no Python process, no dependencies).
The bundled
mcp-server/(a local Python MCP that wraps the REST API) is retained for offline/self-hosted use, but the plugin now uses the hosted MCP by default. To run it locally instead, setmcpServers.webrobotto{ "command": "python3", "args": ["${CLAUDE_PLUGIN_ROOT}/mcp-server/server.py"] }andpip install -r mcp-server/requirements.txt.
Configure credentials
The MCP server reads credentials from, in priority order:
- Environment variables:
WEBROBOT_API_ENDPOINT,WEBROBOT_API_KEY,WEBROBOT_JWT. - Plugin-scoped JSON config:
~/.claude/plugins/webrobot/config.json— preferred when you don't have the CLI installed. - CLI HOCON configs, looked up in this order:
~/.config/webrobot/config.cfg~/.webrobot/config.cfg./config.cfg(current working directory)
Minimal plugin config:
{
"api_endpoint": "https://api.webrobot.eu",
"apikey": "your-api-key:your-secret"
}
apikey is sent as Authorization: ApiKey … and X-API-Key. A jwt
field is also honored and overrides the API key when present.
Default endpoint when nothing is configured: https://api.webrobot.eu.
Skills
| Slash command | What it covers |
|---|---|
/webrobot-overview | Platform overview: vision, architecture, pricing tiers, BYOC model, GTM phases, roadmap, competitive positioning. The "what is this" entry point. |
/webrobot-pipeline | Build, validate, deploy pipeline manifests. Stage catalog, positional args, the LLM-driven e-commerce flow (auto_internal_search → intelligentExplore / intelligentJoin → iextract), and the agentic browser_use stage. |
/webrobot-cli | Use the webrobot CLI to manage projects, jobs, agents, datasets, plugins, pipeline execution. |
/webrobot-admin | Direct admin operations against the WebRobot API: projects, jobs, executions, agents, datasets, cloud credentials, LLM providers. |
/webrobot-plugin-dev | Build WebRobot plugins — custom ETL stages (Scala SDK) and REST API endpoints (Java JAX-RS). |
/webrobot-python-extension | Write, register, and embed Python Extension functions for pipelines without writing Scala. |
/webrobot-frontend-plugin-dev | Build a partner-authored UI plugin for the WebRobot dashboard (Next.js). Plugins ship as a ZIP of ESM bundles, upload to MinIO, hot-load at runtime. |
/webrobot-sdk-java | Consume the WebRobot REST API from JVM apps using the public webrobot-sdk JAR. |
/webrobot-sdk-nodejs | Consume the WebRobot REST API from Node.js / TypeScript using the public npm package. |
/webrobot-sdk-python | Consume the WebRobot REST API from Python using the public package. |
Each skill is a self-contained markdown file under skills/<name>/SKILL.md
plus optional companion files (examples, scripts) in the same directory.
MCP server tools
The server (mcp-server/server.py) exposes the WebRobot REST API as
~30 MCP tools, grouped by domain:
- Auth:
auth_me,auth_check,auth_set. - Projects:
list_projects,get_project,create_project,delete_project. - Jobs:
list_jobs,get_job,create_job,delete_job,execute_job,stop_job. - Executions:
list_executions,get_execution_logs. - Stages (ETL catalog):
list_stages,describe_stage,search_stages,suggest_pipeline_stages. - Manifests:
validate_manifest,apply_manifest,run_pipeline. - Categories + agents:
list_categories,list_agents,get_agent,create_agent,get_agent_code. - Datasets:
list_datasets,delete_dataset. - Cloud credentials:
list_cloud_credentials. - LLM:
list_llm_providers,llm_infer. - Escape hatch:
api_call— call any WebRobot endpoint by path/method when no typed tool fits.
All tools return JSON strings. The MCP server has no business logic of its own; it's a thin, typed wrapper over the REST API + a config loader.
Repository layout
.claude-plugin/plugin.json Plugin manifest read by Claude Code
.mcp.json Stand-alone MCP server entry (for local dev)
skills/
webrobot-overview/ Platform overview skill
webrobot-pipeline/ Pipeline authoring skill
webrobot-cli/ CLI usage skill
webrobot-admin/ Admin operations skill
webrobot-plugin-dev/ ETL stage + REST plugin dev skill
webrobot-python-extension/ Python Extensions skill
webrobot-frontend-plugin-dev/ Dashboard UI plugin skill
webrobot-sdk-java/ Java/JVM SDK skill
webrobot-sdk-nodejs/ Node SDK skill
webrobot-sdk-python/ Python SDK skill
mcp-server/
server.py FastMCP server, ~30 tools
requirements.txt Single dep: mcp>=1.0.0
Development
The plugin has no build step. Local iteration:
- Edit a
skills/<name>/SKILL.mdormcp-server/server.py. - In Claude Code, reload the plugin (
/plugin reload webrobot). - Try the slash command or call the MCP tool.
The MCP server can also be run stand-alone for debugging:
python3 mcp-server/server.py
(Stdio MCP transport; pipe a tools/list request to confirm the tools
load and credentials resolve.)
To add a new skill, create skills/<new-name>/SKILL.md with the
standard frontmatter:
---
name: <slug>
description: <one-line description of when this skill applies>
---
Pull requests welcome — see the WebRobot organization for related repositories (ETL engine, CLI, SDKs, dashboard, portal).
License
MIT. See LICENSE (or .claude-plugin/plugin.json license field).
Maintained by WebRobot Ltd — ceo@webrobot.eu.