mcp-gateway

MCP server gateway for Feel Your Protocol — exposes the mcp-execution-engine to AI agents over stdio (local) and HTTP (later).

Repo: github.com/feelyourprotocol/mcp-gateway · Release: v0.1.0 (stdio, local agents)

Architecture

Agent (Cursor, Claude Desktop, …)
  → MCP stdio transport
    → mcp-gateway (tools + TaskProcessor)
      → mcp-execution-engine (stateless EthereumJS)

Docs: mcp-docs.feelyourprotocol.org

Related MCP server: multivon-mcp

Tools (v0.1.0)

Local development

Clone the execution engine as a sibling directory (required — gateway depends on file:../mcp-execution-engine):

git clone https://github.com/feelyourprotocol/mcp-execution-engine.git
git clone https://github.com/feelyourprotocol/mcp-gateway.git

# From mcp-execution-engine
npm ci && npm run build

# From mcp-gateway
npm ci
npm run test:ci    # build + vitest (30 tests)
npm run build && npm start   # stdio MCP server (for agent config)

Cursor / Claude Desktop

{
  "mcpServers": {
    "feel-your-protocol": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-gateway/dist/index.js"]
    }
  }
}

Build first (npm run build). Use an absolute path to dist/index.js.

After pulling or changing gateway code: run npm run build again, then restart the MCP server in Cursor (Settings → MCP → restart feel-your-protocol, or reload the window). Cursor caches the tool list from the running process — it will not pick up new tools until restart.

On startup the server logs to stderr (visible in MCP logs):

[fyp-mcp] FeelYourProtocol-EVM v0.1.0 ready — tools: describe_capabilities, simulate_evm_bytecode

If you only see describe_capabilities in that line, the running binary is stale.

JSON schemas

Machine-readable tool inputs (also published on mcp-docs):

• schemas/describe_capabilities.input.json

• schemas/simulate_evm_bytecode.input.json

License

MIT