ICARE live esports predictions
RemoteAbout ICARE live esports predictions
ICARE is an MCP server for predicting live esports match outcomes, with markets settled by an AI referee. It serves agents and MCP clients that want match data, predictions, and portfolio tools; an autonomous agent can also place bets within configured limits.
Key features
- 01View live matches and prices
- 02Predict Win or Lose
- 03Check wallet and positions
- 04Sell positions at current price
- 05Run an agent with spending limits
Use cases
- 01Review live match markets
- 02Place a prediction on a match
- 03Monitor positions and portfolio balance
- 04Automate match scanning within limits
MCP support
The local MCP server provides wallet, match, bet, position, sell, and season tools. A read-only remote MCP endpoint offers match data, bet links, season information, search, and fetch.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://app.icarehub.xyz/api/mcpclaude mcp add --transport http djangode-icare https://app.icarehub.xyz/api/mcp{"mcpServers":{"djangode-icare":{"type":"http","url":"https://app.icarehub.xyz/api/mcp"}}}Open in Cursor{"servers":{"djangode-icare":{"type":"http","url":"https://app.icarehub.xyz/api/mcp"}}}Open in VS Code{"mcpServers":{"djangode-icare":{"type":"http","url":"https://app.icarehub.xyz/api/mcp"}}}{"mcpServers":{"djangode-icare":{"serverUrl":"https://app.icarehub.xyz/api/mcp"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
Details
From the project READMEICARE MCP server
Let your AI agent predict live esports games. ICARE follows live streamers (CS2, Valorant, EA SPORTS FC, Rocket League, League of Legends…), opens a Win / Lose market on each game, and an AI referee reads the stream to settle it with a proof published on IPFS and on Base.
- Works with Claude, GPT and Gemini: Claude Desktop, Claude Code, OpenAI Codex CLI, Gemini CLI, Cursor, any MCP client, and claude.ai, ChatGPT and Gemini through the remote server
https://app.icarehub.xyz/api/mcp. - Autonomous agent (
icare-agent): watches live matches, decides with your strategy (built-in rule, Claude, GPT or Gemini), bets within your limits and sells at your profit target, on its own. - Wallet created on your machine on first use (or bring your own key): the key never leaves it.
- Gasless bets over x402: the agent pays its USDC stake with a signed authorization. No ETH, no gas, no approval.
- Sell anytime: ICARE liquidity buys positions back at the current price while the market is open.
- Season 0 (Base Sepolia testnet, test USDC): correct prediction +1, wrong one −1. Leaderboard: https://www.icarehub.xyz/season-0/en
Tools
| Tool | What it does |
|---|---|
icare_wallet | Agent wallet address (created on first use), its test USDC and ICARE Portfolio balances |
icare_live_matches | Open matches with Win/Lose prices, pot, closing time; recent results with their referee proof |
icare_activate | Only when ICARE access codes are on: one-time activation of the agent wallet with an access code |
icare_bet | Predict Win or Lose (1 to 50 USDC). Paid from the agent Portfolio when funded, otherwise over x402 (stake + 0.20 USDC network fees) |
icare_positions | Portfolio balance and positions on recent matches |
icare_sell | Sell a position to ICARE liquidity at the current buyback price |
icare_season | Season 0 leaderboard |
Install
Node.js 22+ and git. One line, no account:
| Client | Command |
|---|---|
| Claude Code | claude mcp add icare -- npx -y github:Djangode/icare-mcp |
| OpenAI Codex CLI | codex mcp add icare -- npx -y github:Djangode/icare-mcp |
| Gemini CLI | gemini mcp add icare npx -y github:Djangode/icare-mcp |
Claude Code plugin: /plugin install icare --marketplace Djangode/icare-mcp (or /plugin marketplace add Djangode/icare-mcp, then /plugin install icare@icare).
Gemini CLI extension: gemini extensions install https://github.com/Djangode/icare-mcp
Claude Desktop, Cursor or any MCP client (JSON configuration):
{
"mcpServers": {
"icare": { "command": "npx", "args": ["-y", "github:Djangode/icare-mcp"] }
}
}
Claude Desktop in one click: download icare.mcpb from the latest release and open it.
Then ask your AI:
- "Show my ICARE wallet": it creates the agent wallet and gives its address.
- Get test USDC for that address at https://faucet.circle.com (Base Sepolia). No ETH needed.
- "Show the live ICARE matches", then "Bet 2 USDC on Win on match 170".
Options: ICARE_AGENT_PRIVATE_KEY (use your own wallet instead), ICARE_STATE_FILE (wallet and session file, default ~/.icare-mcp.json, readable only by you), ICARE_API_URL, ICARE_APP_URL, ICARE_RPC_URL.
Access codes: none needed while accessRequired is false in https://api.icarehub.xyz/api/agents/config. When ICARE turns them on, codes are given in the ICARE Discord, https://discord.gg/SmRynyJmnK, and icare_activate is called once.
claude.ai, ChatGPT and Gemini (remote server)
Add a custom connector with the URL https://app.icarehub.xyz/api/mcp (Streamable HTTP, no sign-in):
- claude.ai: Settings → Connectors → Add custom connector.
- ChatGPT: turn on developer mode in the connector settings, then create a connector with this URL.
- Gemini CLI without a local server:
gemini mcp add --transport http icare https://app.icarehub.xyz/api/mcp
The remote server is read-only and never holds a key: icare_live_matches, icare_match, icare_bet_link (the match page where you predict with your own wallet), icare_season, plus search and fetch. For automatic predictions, run the agent below.
Autonomous agent
npx -y -p github:Djangode/icare-mcp icare-agent run --stake 1 --daily 5
It uses the same wallet as icare-mcp (created on first use), scans the live matches every minute, and bets only within your limits.
| Option | Default | Meaning |
|---|---|---|
--brain rules | yes | Built-in rule, no AI: bets when the streamer's ICARE record beats the price by --edge |
--brain claude | Asks Claude through Claude Code (claude -p) | |
--brain command --brain-cmd "codex exec" | Any CLI that takes the prompt last: GPT (codex exec), Gemini (gemini -p) | |
--strategy "..." | Your strategy in plain English, for the AI brains | |
--stake 1 / --max-stake 5 | 1 / 5 | Default and highest stake in USDC (1 to 50) |
--daily 10 | 10 | USDC spent at most over 24 h, network fees included |
--max-open 3 | 3 | Positions open at the same time |
--take-profit 0.3 | 0.3 | Sells when the buyback pays 30% over the stake (0 = hold to the result) |
icare-agent once runs a single scan; icare-agent status shows the wallet, settings, positions and last actions. Memory and journal: ~/.icare-agent.
How a bet works
- The agent signs the Portfolio action that places its prediction on the ICARE Match contract.
POST https://api.icarehub.xyz/api/agents/betsanswers 402 Payment Required (x402 v2, schemeexact, USDC oneip155:84532).- The agent signs an ERC-3009
transferWithAuthorizationfor the stake + 0.20 USDC and retries withPAYMENT-SIGNATURE. - ICARE collects it, credits the agent Portfolio and relays the signed action (about 40 seconds: three confirmed transactions). If the market closed in between, the stake stays on the agent Portfolio.
Testnet only for now: test USDC has no value. 18+. Availability of paid skill competitions varies by jurisdiction.
MIT licence.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.