crypto_mcp
Provides cryptocurrency market data (prices, symbols) and a gated demo trading flow via MCP tools, with safety defaults like dry-run and confirmation gates.
README
crypto_mcp
crypto_mcp is a refactor of crypto_bot into a single MCP-first repository.
Python requirement: >=3.11.
For user-facing capability map, see: docs/CAPABILITIES.MD.
What this project does
- Expose deterministic MCP tools over stdio.
- Keep transport contract stable for LLM clients.
- Separate market data (safe/read-only) from trading (gated).
- Keep exchange adapters replaceable.
Architecture
src/crypto_mcp/mcp_server.py: FastMCP server and tool definitions.src/crypto_mcp/agent.py: NL command router mapping prompts to MCP tools.src/crypto_mcp/config.py: Typed settings loader with sane defaults and clamps.src/crypto_mcp/adapters/binance.py: Binance HTTP adapter abstraction.src/crypto_mcp/main.py: Runnable MCP entrypoint.tests/: MCP contract, core server, and agent unit tests.
Detailed architecture: docs/ARCHITECTURE.MD.
Capabilities quick table
| Area | What it does | Main MCP tools |
|---|---|---|
| Runtime health | Reports service state, limits, mode, exchanges | health |
| Exchange inventory | Lists adapter-backed exchanges | list_exchanges |
| Market data | Fetches ticker prices and symbol lists | get_price, list_exchange_symbols |
| Guarded execution | Submits demo orders with explicit confirmation gate | submit_demo_order, confirm_pending_order |
| NL routing | Maps natural language to MCP tool calls | crypto-mcp-agent |
Design pattern references taken from polymarket_mcp:
- deterministic tool envelope:
{ok, tool, ...} - bounded limits for list/read tools
- safety confirmation flow for trade tools
- split entrypoint (
main.py) from MCP server implementation
Quickstart
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
python -m pytest
crypto-mcp-server
# separate terminal
crypto-mcp-agent
Setup path-style variant:
cd /path/to/crypto_mcp
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
Alternative module runner:
python -m crypto_mcp.main
crypto-mcp
Safety defaults
DRY_RUN=true
Default path is non-live demo execution.
Run MCP server locally
Stdio server (local MCP clients):
crypto-mcp-server
For MCP Inspector debugging:
npx @modelcontextprotocol/inspector python -m crypto_mcp.main
Do not write print() logs inside stdio server paths. Use logging to stderr.
Available MCP tools:
healthlist_exchangesget_pricelist_exchange_symbolssubmit_demo_orderconfirm_pending_order
Tool response envelope is consistent:
ok: booleantool: tool nameerror+error_categoryon failures- tool-specific payload under
payload
MCP limits are controlled by:
MCP_DEFAULT_LIMITMCP_MAX_LIMITMCP_CONTEXT_MODE(sharedorrequest)
Order safety controls are controlled by:
DRY_RUNREQUIRE_CONFIRMATION_ABOVE_USD
Connect this MCP to clients
Use path placeholders:
path/to/crypto_mcp= repository root.path/to/crypto_mcp/.venv/bin/python= project interpreter.
GitHub Copilot (VS Code Agent Mode)
Add MCP server config in your VS Code MCP settings JSON:
{
"mcpServers": {
"crypto-mcp": {
"command": "python",
"args": ["-m", "crypto_mcp.main"],
"cwd": "path/to/crypto_mcp",
"env": {
"PYTHONPATH": "src"
}
}
}
}
If your environment requires venv binary, use:
{
"command": "path/to/crypto_mcp/.venv/bin/python",
"args": ["-m", "crypto_mcp.main"]
}
Claude Desktop
Edit Claude Desktop MCP config and add:
{
"mcpServers": {
"crypto-mcp": {
"command": "path/to/crypto_mcp/.venv/bin/python",
"args": ["-m", "crypto_mcp.main"],
"cwd": "path/to/crypto_mcp",
"env": {
"PYTHONPATH": "src"
}
}
}
}
Restart Claude Desktop after changes.
OpenCode
OpenCode can run MCP from command config. Add server entry in your OpenCode MCP configuration:
{
"mcpServers": {
"crypto-mcp": {
"command": "path/to/crypto_mcp/.venv/bin/python",
"args": ["-m", "crypto_mcp.main"],
"cwd": "path/to/crypto_mcp",
"env": {
"PYTHONPATH": "src"
}
}
}
}
Then restart OpenCode session and verify tools:
healthlist_exchangesget_pricelist_exchange_symbolssubmit_demo_orderconfirm_pending_order
OpenCode install and usage:
# install OpenCode (example)
npm install -g opencode-ai
# verify installation
opencode --version
# start session in this repository
cd path/to/crypto_mcp
opencode
Inside OpenCode:
- ensure MCP server
crypto-mcpis connected. - run a quick tool check with
health. - ask natural language command examples:
- "get btcusdt price"
- "list 20 symbols"
- "buy 50 btcusdt"
Environment
Copy .env.example to .env and update values.
MCP_DEFAULT_LIMIT,MCP_MAX_LIMITMCP_CONTEXT_MODE(sharedorrequest)DRY_RUN(true/false)EXCHANGES_ENABLED(binancedefault)GEMINI_API_KEY(optional, used by local NL routing layer)BINANCE_API_BASE_URLBINANCE_API_KEY,BINANCE_API_SECRETREQUIRE_CONFIRMATION_ABOVE_USD
For this task, an empty .env file exists and is safe because defaults are clamped.
What this MCP can do
Read-only tools:
health: service status, context mode, limits, enabled exchanges.list_exchanges: list adapter-backed exchanges.get_price: fetch ticker price for symbol/exchange.list_exchange_symbols: bounded symbol list with clamp.
Trading flow tools (safety-gated):
submit_demo_order: validates side/symbol/size and creates pending confirmation above threshold.confirm_pending_order: executes pending order if valid and not expired.
Order confirmation flow:
- Call
submit_demo_order. - If payload has
status=pending_confirmation, captureconfirmation_id. - Call
confirm_pending_order(confirmation_id).
Contract:
- deterministic response envelope:
{ok, tool, payload}on success. - deterministic error envelope:
{ok, tool, error, error_category}on failure. - live trading intentionally disabled: returns
live_trading_not_implementedwhenDRY_RUN=false.
Safety Notes
submit_demo_ordernever executes live orders; withDRY_RUN=falseit returnslive_trading_not_implemented.submit_demo_orderrequires explicit confirmation for large notional.- response shape never raises raw exceptions to MCP caller; errors are categorized.
Tests
python -m pytest
Important notes
- Local MCP stdio transport must keep stdout clean JSON-RPC only.
- This project is starter infrastructure. Not trading advice.
- Add stronger risk modules before real funds.
Known improvement areas (scan summary)
- Add adapter-level symbol tradability validation before order acceptance.
- Replace broad internal exception mapping with sanitized provider error codes.
- Add confirmation-expiry regression tests at MCP wrapper level.
- Add optional HTTP transport runner for remote clients.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.