Plzebo runs on your own machine — the client starts it, so there is no endpoint to ping. 52 installs a week from npm. Last commit 14 Aug 2026.
Cost-parity bench for agent loops: $/success and time across model mixes, with verifiable receipts.
Today is the operative word: we check Plzebo 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 plzebo -- npx -y plzebo
{
"mcpServers": {
"plzebo": {
"command": "npx",
"args": [
"-y",
"plzebo"
]
}
}
}
[mcp_servers.plzebo]
command = "npx"
args = ["-y", "plzebo"]
{
"mcpServers": {
"plzebo": {
"command": "npx",
"args": [
"-y",
"plzebo"
]
}
}
}
{
"servers": {
"plzebo": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"plzebo"
]
}
}
}
Optional settings from the registry entry, left out of the block above:
ANTHROPIC_API_KEY (Real Anthropic key for placebo-mode arms; arms only ever see placebo keys. Optional when every arm uses passthrough auth), OPENAI_API_KEY (Real OpenAI-compatible key for placebo-mode arms. Optional when every arm uses passthrough auth).
Human-in-the-loop approval for agent actions, with verifiable action-bound receipts.
Pay-per-call APIs and MCP services for agents, no accounts or keys, with verifiable receipts.
Local MCP runtime for policy-governed, verified agent actions with durable receipts.
Real-time, drift-verified LLM pricing across 31 providers with cost+quality model routing.
Agentic rails for complex workflows with receipts, fees, and MCP tool access.
Deterministic claim verification with receipts across ~60 domains. No model in the loop.
Agent-payable open-loop tracker for refunds, returns, and asks with TRACE receipts.
Compare prices across French merchants, with recorded price history and verified price drops.
Answers built from our own checks of this server.