Agent Economy Index · MCP census · probed 2026-09-29 · JSON

com.completionkit/evals

Verdict: 5 tool description(s) carry model-directed, hidden, secret-material or exfiltration patterns. Endpoint https://completionkit.com/mcp. Latency 10,868 ms.

Tools (54 fingerprinted)

tooldescription lengthflagsfirst seenlast seen
agreements_create1702026-09-292026-09-29
agreements_list732026-09-292026-09-29
datasets_create4332026-09-292026-09-29
datasets_create_from_url4422026-09-292026-09-29
datasets_delete162026-09-292026-09-29
datasets_get192026-09-292026-09-29
datasets_list172026-09-292026-09-29
datasets_update162026-09-292026-09-29
judges_compare2062026-09-292026-09-29
judges_replay2692026-09-292026-09-29
metric_groups_create212026-09-292026-09-29
metric_groups_delete212026-09-292026-09-29
metric_groups_get242026-09-292026-09-29
metric_groups_list222026-09-292026-09-29
metric_groups_update212026-09-292026-09-29
metric_versions_dismiss1942026-09-292026-09-29
metric_versions_list1512026-09-292026-09-29
metric_versions_publish2982026-09-292026-09-29
metrics_create5782026-09-292026-09-29
metrics_delete152026-09-292026-09-29
metrics_get182026-09-292026-09-29
metrics_list162026-09-292026-09-29
metrics_suggest_variants3192026-09-292026-09-29
metrics_update5532026-09-292026-09-29
promptfoo_import3132026-09-292026-09-29
prompts_create152026-09-292026-09-29
prompts_delete152026-09-292026-09-29
prompts_get182026-09-292026-09-29
prompts_list162026-09-292026-09-29
prompts_publish552026-09-292026-09-29
prompts_suggest_improvement3122026-09-292026-09-29
prompts_update2692026-09-292026-09-29
provider_credentials_create28secret_material2026-09-292026-09-29
provider_credentials_delete28secret_material2026-09-292026-09-29
provider_credentials_get56secret_material2026-09-292026-09-29
provider_credentials_list56secret_material2026-09-292026-09-29
provider_credentials_update28secret_material2026-09-292026-09-29
responses_get232026-09-292026-09-29
responses_list3602026-09-292026-09-29
runs_create1532026-09-292026-09-29
runs_delete122026-09-292026-09-29
runs_generate1882026-09-292026-09-29
runs_get2562026-09-292026-09-29
runs_list132026-09-292026-09-29
runs_regrade1622026-09-292026-09-29
runs_rerun1722026-09-292026-09-29
runs_retry_failures982026-09-292026-09-29
runs_update122026-09-292026-09-29
tags_create372026-09-292026-09-29
tags_delete812026-09-292026-09-29
tags_get152026-09-292026-09-29
tags_list132026-09-292026-09-29
tags_update132026-09-292026-09-29
usage_get3542026-09-292026-09-29

Tool changes (0)

whentoolkind
none yet

Probe history

2026-09-29 ok (54 tools)

Depend on this server? Watch it with ToolDrift, $29/month: an alert within six hours when it changes a tool or trips a detector. Flags here are pattern matches on tool descriptions, not verdicts; read the description before deciding.
Run this server? Show today's check on your README or site with a badge that refreshes daily: badge
<a href="https://index.agentexchange.work/mcp-server/com.completionkit%2Fevals"><img src="https://index.agentexchange.work/badge/mcp/com.completionkit%2Fevals.svg" alt="MCP server check by agentexchange.work"></a>
Claim this server — prove you operate completionkit.com and this page gets a "Claimed by the operator" line, the badge gets a CLAIMED prefix, and you get a JSON webhook alert when a daily probe records a tool change or more flagged descriptions than the day before (at most one alert per 6 hours). No account, no email, nothing to install.
  1. Put a text file at https://completionkit.com/.well-known/agentexchange-claim.txt whose entire content is exactly com.completionkit/evals.
  2. Submit this form. We fetch that URL once (7-second timeout, User-Agent agentexchange-index-claim) and compare the trimmed body with the registry name.

Alert payload: {event, server, day, drift:[{tool, kind, detected_at}], findings, previous_findings, page_url, badge_url}. Public list of claims (names and hosts only): /v1/claims.json.

Run this server? MCP servers process users' data and often accept content; the laws that follow are a 45-day privacy request process, an opt-out path, and for hosted content a takedown or notice-and-action process. Check your site free or get a hosted desk.