One MCP endpoint. Twelve tools.
A native Model Context Protocol server you connect in one line - point your agent at our managed endpoint, or self-host with Docker. Every Ujeebu endpoint exposed as a typed tool, ready to plug into Claude Desktop, Cursor, Windsurf, or any MCP-compliant agent. Bring your API key, get web access in 60 seconds.
# claude_desktop_config.json
{
"mcpServers": {
"ujeebu": {
"url": "https://mcpdev.ujeebu.com/mcp",
"headers": { "Authorization": "Bearer uj_live_..." }
}
}
}
# In Claude:
> Find me the top 5 React component
> libraries by GitHub stars and
> summarize their differences.
🛠 ujeebu.serp({ q: "react component library" })
🛠 ujeebu.scrape(url: github.com/...) ×5
🛠 ujeebu.extract(url: ...) ×5Real Model Context Protocol, not a wrapper, not a translation layer. Tools, resources, and prompts conform to the spec.
scrape (html, screenshot, pdf, json, by-rules), extract, card, serp, markdown, chatgpt, gemini, account. Each ships with full JSON Schema so the agent picks correctly.
Cap how many credits an agent can burn per session. Hard stop at the limit. No nightmare bills from a rogue eval loop.
Every tool result includes the source URL, fetch timestamp, and credit cost. Your agent can ground its claims; you can audit them.
Point your agent at our managed endpoint - zero install, always current - or run it yourself from the ujeebu/ujeebu-mcp Docker image, inside your own network or CI.
Works with Claude Desktop, Cursor, VS Code, Windsurf, the OpenAI Agents SDK, Google ADK, and any other MCP-compatible host. Same config, every client.
Paste the config. Restart. Done.
No npx, no Homebrew, nothing to compile. Point your host at our hosted endpoint, or self-host with Docker. Every host block below adapts to your choice.
We run the server for you at https://mcpdev.ujeebu.com/mcp - nothing to install. The host blocks below authenticate with your API key.
# Self-host from the published image (Docker Hub) - your MCP host launches it:
docker run -i --rm -e UJEEBU_KEY=YOUR_KEY ujeebu/ujeebu-mcpRequires Docker on the host. The blocks below switch to the Docker command automatically.
Stays in your browser. Every snippet below regenerates as you type.
~/Library/Application Support/Claude/claude_desktop_config.json
macOS
%APPDATA%\Claude\claude_desktop_config.json
Windows
Then quit and reopen Claude Desktop.
~/.cursor/mcp.json
global
.cursor/mcp.json
per-project
Or paste into the config file and reload Cursor.
~/.vscode/mcp.json
user
.vscode/mcp.json
workspace
Requires VS Code 1.95+ with MCP support enabled.
~/.codeium/windsurf/mcp_config.json
Then reload Windsurf (Cmd-Shift-P → Reload Window).
pip install openai-agents
Tools auto-discovered from the MCP server:
scrape, serp, extract, and more.
Works with any OpenAI-compatible model - DeepSeek, Kimi, Qwen, Groq, vLLM, Ollama. Just point an AsyncOpenAI client at the provider's base_url and pass it to OpenAIChatCompletionsModel.
pip install google-adk
Add the toolset to any ADK Agent. Works with Gemini, Vertex, or self-hosted models.
Hosted: connect over streamable HTTP with an Authorization: Bearer header.
Docker: launch the stdio command with UJEEBU_KEY in env.
Stuck? Read the MCP setup guide or ping support.
Real things real teams shipped this quarter.
A junior researcher asks Claude Desktop to compare four GitHub libraries. The agent runs SERP → article → markdown → reasons over the result. 90 seconds end-to-end.
In Cursor, ask "what does Stripe's API say about idempotency keys?" The MCP server fetches, summarizes, and pastes the source URL into your editor.
Build a vertical agent (legal research, sales prospecting). The MCP server gives it instant web access; you focus on the domain logic.
Wire MCP into your eval framework. Per-key budgets stop runaway test loops; full request logs let you reproduce any agent step exactly.
Ship MCP Server tonight.
5,000 credits free. No card. Real residential proxies on the free tier.