BPE MCP Server

Model Context Protocol server for the BPE market-data API. Lets Claude Desktop, Cursor, Goose, and other MCP-enabled agents query consolidated Bitcoin pricing, ML signals, funding-rate skew, and natural-language market briefings — without writing any HTTP boilerplate.

Status: v0.2 — five tools, local stdio transport, no LLM calls inside the briefing endpoint (template-rendered). API key required.

Listed in the official MCP Server Registry as io.github.bxetech/bpe-mcp. Install via npm install -g @bxetech/bpe-mcp or npx -y @bxetech/bpe-mcp.

Why use it

If you're building an agent that touches Bitcoin markets, you have two options today:

1. Talk to 36 different exchange APIs directly. Each has a different schema, different auth, different rate limits, different reliability. Your agent burns tokens reasoning over inconsistent payloads.

2. Use one consolidated source. BPE aggregates all 36 venues into a single normalised feed with derived signals (ML predictions, sentiment, funding skew, basis). One auth, one schema, one tool call.

This MCP server is option 2 dressed up as native tools your agent can discover and call.

Related MCP server: Backtesting Arena

Tools (v0.2)

Install

npm install -g @bxetech/bpe-mcp

Or use npx with no install (recommended for Claude Desktop):

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "bpe": {
      "command": "npx",
      "args": ["-y", "@bxetech/bpe-mcp"],
      "env": {
        "BPE_API_KEY": "bpe_..."
      }
    }
  }
}

Restart Claude Desktop. The five tools appear under the MCP server picker.

Configuration

Local development

git clone https://github.com/bxetech/bpe-mcp
cd bpe-mcp
npm install
npm run build
BPE_API_KEY=... npm run dev

To dogfood in Claude Desktop while iterating, point at the local build:

{
  "mcpServers": {
    "bpe-dev": {
      "command": "node",
      "args": ["/abs/path/to/mcp-server/dist/index.js"],
      "env": {
        "BPE_API_KEY": "...",
        "BPE_BASE_URL": "http://localhost:8082"
      }
    }
  }
}

Roadmap

• v0.2 — get_sentiment_snapshot ✅, friendly tier-aware errors ✅, timestamped briefings ✅

• v0.3 — get_basis, subscribe_alert (webhook subscription), hosted HTTP transport for non-developer install (Cursor extension, ChatGPT Desktop)

• v0.4 — Optional LLM-rendered briefing via Haiku (richer prose, with per-call cost)

See docs/research/AGENT_INTEGRATION_AND_MONETISATION_2026-04-27.md for the broader product / monetisation plan.

License

MIT — see LICENSE.