Agent Referee runs on your own machine — the client starts it, so there is no endpoint to ping. 989 installs a week from npm. Last commit 30 Oct 2026.
Neutral tie-breaks between AI agents and sealed review of an agent's output, with signed receipts.
Today is the operative word: we check Agent Referee every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add agent-referee -- npx -y agent-referee-mcp
{
"mcpServers": {
"agent-referee": {
"command": "npx",
"args": [
"-y",
"agent-referee-mcp"
]
}
}
}
[mcp_servers.agent-referee]
command = "npx"
args = ["-y", "agent-referee-mcp"]
{
"mcpServers": {
"agent-referee": {
"command": "npx",
"args": [
"-y",
"agent-referee-mcp"
]
}
}
}
{
"servers": {
"agent-referee": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"agent-referee-mcp"
]
}
}
}
Optional settings from the registry entry, left out of the block above:
AGENT_REFEREE_PRIVATE_KEY (Optional. Private key of a dedicated Base wallet holding a few dollars of USDC, for paid sealed reviews (also start the server with the x402-fetch package: npx -y -p agent-referee-mcp -p x402-fetch agent-referee-mcp). Free tools work without it), AGENT_REFEREE_DAILY_LIMIT_USD (Optional. Most this server may spend per UTC day (default 1.00; 0 turns paid calls off)).
Signed, sealed handoffs between AI agents — trust computed by the receiver.
Signed, append-only receipts of an AI agent's decisions and outcomes, with public proof links.
Score documents and check agent work — signed PactCheck Receipts: proof-of-review for agents.
Owner-signed permissions for AI agent actions; ALLOW/BLOCK/REVIEW with request-bound receipts.
Marketplace and payment rail for AI agents: list, buy and settle with signed receipts.
Durable symmetric deliberation between two AI agents, with isolated worktrees and peer review.
Durable symmetric deliberation between two AI agents, with isolated worktrees and peer review.
Independent pass/fail/retry judge and false-consensus check for agent outputs, with signed receipts.
Answers built from our own checks of this server.