Zipp MCP

Multi-language crypto news for AI assistants — editorial summaries, sentiment labels (BULLISH / NEUTRAL / BEARISH), importance scores (0–100), and every story credited to its original publisher.

This repository hosts the public listing manifest and the zipp-mcp Python package. The canonical Zipp MCP server is hosted at https://zippfeed.com/mcp/ — most clients should connect to that URL directly. The PyPI / Docker package is for the cases where they can't: stdio-only desktop clients, locked-down networks, or anywhere you want a self-contained install.

• 🌐 Website: zippfeed.com
• 🔌 MCP endpoint: https://zippfeed.com/mcp/
• 📡 Transport: Streamable HTTP (MCP spec 2025-06-18)
• 🔓 Auth: none — public, rate-limited at the Cloudflare edge
• 📰 Coverage: crypto / blockchain / Web3 across 8 languages
• ✍️ Editorial: every item carries sentiment + importance + source attribution
• 📦 PyPI: uvx zipp-mcp · Docker: ghcr.io/deficlow/zipp-mcp (Day 3 follow-up)

Self-host

Three install paths. All three speak the same protocol and call the same upstream API; the right choice depends on the client.

uvx — no install, one command

For stdio MCP clients (Claude Desktop, Cursor's stdio mode, Cline, Zed, etc.) that prefer launching a local subprocess:

uvx zipp-mcp

Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "zipp": {
      "command": "uvx",
      "args": ["zipp-mcp"]
    }
  }
}

pip install — for embedded use

pip install zipp-mcp
zipp-mcp --transport stdio

Or import the client directly in your own code:

from zipp_mcp import ZippClient

async with ZippClient() as zipp:
    payload = await zipp.search(query="bitcoin etf", lang="en-US", limit=5)
    for item in payload["items"]:
        print(item["title"], item["sentiment"], item["importance"])

Docker — for hosted / sandboxed environments

docker run -i --rm ghcr.io/deficlow/zipp-mcp

For Streamable HTTP transport on a server (Railway, Fly, Render):

docker run -p 8080:8080 -e MCP_HOST=0.0.0.0 \
  ghcr.io/deficlow/zipp-mcp \
  zipp-mcp --transport http

Configuration

All flags read from env vars too; everything is optional.

The package is read-only — there is nothing to authenticate; the upstream API is public and rate-limited at the Cloudflare edge.

Why Zipp?

Most news connectors are pure aggregators that hand the model a raw headline list. Zipp adds an editorial layer on top of the firehose:

Quick start

Zipp speaks Streamable HTTP at https://zippfeed.com/mcp/ — no auth, no install. Below are copy-paste configs for the major MCP-capable clients, plus two ways to smoke-test the endpoint.

Claude Desktop

Add to your claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "zipp": {
      "type": "http",
      "url": "https://zippfeed.com/mcp/"
    }
  }
}

Restart Claude Desktop. You should see Zipp in the tool list and be able to ask things like "What's breaking in crypto right now?" or "Search Zipp for Bitcoin ETF inflows in Turkish."

Claude Code (CLI)

One-liner — registers Zipp at user scope so every project sees it:

claude mcp add --scope user --transport http zipp https://zippfeed.com/mcp/

Verify with claude mcp list.

Claude.ai web (Connectors)

Settings → Connectors → Add custom connector → URL: https://zippfeed.com/mcp/ → no authentication.

ChatGPT (Custom Connectors)

Settings → Connectors → Add → Custom connector → URL: https://zippfeed.com/mcp/ → Authentication: None.

Cursor

~/.cursor/mcp.json:

{
  "mcpServers": {
    "zipp": {
      "url": "https://zippfeed.com/mcp/"
    }
  }
}

Windsurf

~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "zipp": {
      "serverUrl": "https://zippfeed.com/mcp/"
    }
  }
}

Cline (VS Code extension)

Open the Cline panel → MCP Servers → Edit Configuration, then add Zipp to cline_mcp_settings.json:

{
  "mcpServers": {
    "zipp": {
      "url": "https://zippfeed.com/mcp/",
      "type": "streamableHttp"
    }
  }
}

Zed

~/.config/zed/settings.json — Zed's context_servers interface is stdio-only today, so use the mcp-remote bridge:

{
  "context_servers": {
    "zipp": {
      "command": {
        "path": "npx",
        "args": ["-y", "mcp-remote", "https://zippfeed.com/mcp/"]
      }
    }
  }
}

Gemini CLI

~/.gemini/settings.json:

{
  "mcpServers": {
    "zipp": {
      "httpUrl": "https://zippfeed.com/mcp/"
    }
  }
}

OpenAI Responses API

from openai import OpenAI

client = OpenAI()
resp = client.responses.create(
    model="gpt-5",
    tools=[{
        "type": "mcp",
        "server_label": "zipp",
        "server_url": "https://zippfeed.com/mcp/",
        "require_approval": "never",
    }],
    input="What's breaking in crypto right now? Cite the original publisher.",
)
print(resp.output_text)

Anthropic Messages API

The MCP connector is a beta — pass the header below until it goes GA:

import anthropic

client = anthropic.Anthropic()
msg = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    mcp_servers=[{
        "type": "url",
        "url": "https://zippfeed.com/mcp/",
        "name": "zipp",
    }],
    messages=[{"role": "user", "content": "What's breaking in crypto right now?"}],
    extra_headers={"anthropic-beta": "mcp-client-2025-04-04"},
)
print(msg.content)

Smoke test

Verify the server from your shell — no client needed:

curl -s -X POST "https://zippfeed.com/mcp/" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"smoke","version":"1.0"}}}'

Expected: a JSON-RPC initialize result where serverInfo.name = "Zipp" and the server advertises tools, resources, and prompts capabilities.

For an interactive UI, run the official MCP Inspector:

npx @modelcontextprotocol/inspector

In the browser tab that opens, pick Streamable HTTP as the transport, paste https://zippfeed.com/mcp/, click Connect, then browse the Tools / Resources / Prompts tabs to exercise the server end-to-end.

Tools

lang defaults to en-US. See Languages below for the full list.

Sample response

get_latest(lang="en-US", limit=1) returns shape: