CompletionKit
com.completionkit/evals · v1.0.0 · score 89/100 · checked 2026-11-22
Automated, read-only signal from public data, not an audit or an accusation. Maintainers can dispute any finding. How grading works
Prompt evals over MCP: run a prompt on your dataset, score each output 1-5 with an LLM judge.
Source: https://github.com/homemade-software-inc/completion-kit
Findings
| Severity | Rule | Detail |
|---|---|---|
| Low | MW005 | Remote endpoint lists 52 tool(s) to anyone, no credentials needed. |
| Medium | MW102 | 7 tool(s) can execute commands or delete data (prompts_delete, runs_delete, datasets_delete, metrics_delete, metric_groups_delete and 2 more). Require confirmation before the agent calls them. |
What was checked
- https://completionkit.com/mcp: answers without auth
Public tools: prompts_list, prompts_get, prompts_create, prompts_update, prompts_delete, prompts_publish, prompts_suggest_improvement, runs_list, runs_get, runs_create, runs_update, runs_delete, runs_generate, runs_regrade, runs_rerun, runs_retry_failures, responses_list, responses_get, datasets_list, datasets_get, datasets_create, datasets_update, datasets_delete, datasets_create_from_url, metrics_list, metrics_get, metrics_create, metrics_update, metrics_delete, metrics_suggest_variants
Maintainer? Add the badge
Grades refresh daily. Think a finding is wrong? Dispute this grade.
[](https://www.mcpwatch.io/s/com-completionkit-evals-d9739064)