icemcp

MCP (Model Context Protocol) server for the ICEvent ecosystem — exposes OneBlock and Defunds canister services on the Internet Computer to any MCP-compatible AI client.

---

Features

OneBlock tools

Tool	Description
oneblock_get_profile	Fetch a life-path profile by ID
oneblock_get_profile_by_principal	Look up a profile by IC principal
oneblock_search_profiles	Search profiles by display name
oneblock_list_profiles	Paginated list of all profiles
oneblock_get_profile_count	Total number of registered profiles
oneblock_get_default_profiles	Featured / showcase profiles

Defunds tools

Tool	Description
defunds_get_all_grants	All grant proposals
defunds_get_grant	Single grant by ID
defunds_get_grant_voting_status	Voting status for a grant
defunds_get_grant_comments	Community comments on a grant
defunds_get_total_donations	Total accumulated donations
defunds_get_total_voting_power	Total accumulated voting power
defunds_get_public_groups	Public community group funds
defunds_get_group	Single group by ID
defunds_get_group_proposals	Proposals for a specific group
defunds_get_all_proposals	All proposals across all groups
defunds_get_public_ai_agent_funds	Public AI-agent managed funds
defunds_get_voting_policy	Global voting policy parameters
defunds_get_exchange_rates	Current donation exchange rates

---

Requirements

• Node.js ≥ 18
• npm ≥ 9

Installation

npm install
npm run build

Usage

Stdio transport (recommended for AI clients)

Run the MCP server directly; it communicates over stdin / stdout:

node dist/index.js

Claude Desktop integration

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "icemcp": {
      "command": "node",
      "args": ["/path/to/icemcp/dist/index.js"]
    }
  }
}

npx (after publishing to npm)

{
  "mcpServers": {
    "icemcp": {
      "command": "npx",
      "args": ["-y", "icemcp"]
    }
  }
}

---

Deploying on Vercel

The repository includes a Vercel-ready HTTP handler (api/mcp.ts) that uses the MCP Streamable HTTP transport in stateless mode — perfect for serverless functions.

One-click deploy

Manual deploy

npm install -g vercel
vercel

The MCP endpoint will be available at:

https://<your-deployment>.vercel.app/mcp

Connecting an AI client to the Vercel deployment

Use the url transport in your MCP client configuration:

{
  "mcpServers": {
    "icemcp": {
      "url": "https://<your-deployment>.vercel.app/mcp"
    }
  }
}

---

Architecture

The server uses:

• @modelcontextprotocol/sdk — MCP server framework
• @dfinity/agent — Internet Computer HTTP agent
• @dfinity/candid — Candid IDL encoding / decoding

All canister calls are anonymous query calls — no wallet or identity is required. Only publicly readable canister methods are exposed.

Canister IDs (IC mainnet)

Service	Canister ID
OneBlock profile	nzxho-uqaaa-aaaak-adwxq-cai
Defunds backend	ixuio-siaaa-aaaam-qacxq-cai