Evals MCP Server runs on your own machine — the client starts it, so there is no endpoint to ping. 47 installs a week from npm. Last commit 22 Aug 2026.
Author verifiable eval records through a draft→review→revise→submit loop with enforced graders.
Today is the operative word: we check Evals MCP Server 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 evals-mcp-server -- npx -y @cyanheads/evals-mcp-server
{
"mcpServers": {
"evals-mcp-server": {
"args": [
"-y",
"@cyanheads/evals-mcp-server"
],
"command": "npx"
}
}
}
[mcp_servers.evals-mcp-server]
command = "npx"
args = ["-y", "@cyanheads/evals-mcp-server"]
{
"mcpServers": {
"evals-mcp-server": {
"args": [
"-y",
"@cyanheads/evals-mcp-server"
],
"command": "npx"
}
}
}
{
"mcpServers": {
"evals-mcp-server": {
"args": [
"-y",
"@cyanheads/evals-mcp-server"
],
"command": "npx"
}
}
}
This one needs environment variables set before it will start:
MCP_HTTP_HOST (The hostname for the HTTP server.), MCP_HTTP_PORT (The port to run the HTTP server on.), MCP_HTTP_ENDPOINT_PATH (The endpoint path for the MCP server.), MCP_AUTH_MODE (Authentication mode to use: 'none', 'jwt', or 'oauth'.), EVALS_DATA_DIR (Root folder for record JSON. The store manages drafts/, submitted/, and exports/ subdirs under it.), EVALS_REQUIRE_CONFIRMATION (When 'true', evals_submit_draft requests human confirmation before finalizing.), EVALS_DEFAULT_LICENSE (Default metadata.license applied when a draft omits one.), EVALS_CAPTURE_DIR (Directory of framework-written tool-call captures; when set, captures EvalsIDs resolve to full dumps.), MCP_LOG_LEVEL (Sets the minimum log level for output (e.g., 'debug', 'info', 'warn').).
The author declared them in the registry entry; get the values from the project itself.
Human-in-the-loop approval for agent actions, with verifiable action-bound receipts.
The first MCP governed by ICA. Every action is gated and sealed with a verifiable receipt.
Human-in-the-loop for AI agents over MCP: durable approvals with a hosted review page & audit trail
US liquor license compliance records (CA, TX, NY, FL, IL) with verifiable provenance.
Get matched with a top-rated, review-verified real estate agent in the US or Canada.
Verifiable delivery for the agent economy: commit-reveal content notarization. Sellers commit to a d
Verifiable action logs for AI agents. Every action is recorded in a SHA-256 hash chain.
Your private media library and records, with receipts: search, recordings, and a verifiable ledger.
Answers built from our own checks of this server.