# CompletionKit (completionkit.com)

Runs a prompt template over a dataset across several LLM providers, then scores the outputs with an LLM judge against your metrics.

- Atlas card: https://inite.ai/en/atlas/completionkit.com
- Site: https://completionkit.com
- For: developers and teams building AI apps who need to measure prompt changes before shipping
- Categories: data
- Use 0/100 · alive 79/100

## Used for

- score prompt outputs against custom metrics using an LLM judge
- compare prompt versions across models on the same dataset
- generate revised prompts from judge feedback on your data

## Price

- from 29 USD/month, free plan (read 2026-10-21 on https://completionkit.com/pricing)

## For developers

- API docs: https://completionkit.com/docs/api
- MCP server: com.completionkit/evals (namespace; open on 2026-10-21, 54 tools)
- Status page: https://status.completionkit.com
- Read 2026-10-21

## MCP server, measured daily

Server: https://completionkit.com/mcp, checked every day since 2026-10-21 (4 checks).

- Answered: measuring, 4 daily checks so far
- Median first answer: 316 ms
- Sign-in: none
- Context its tool definitions take: about 6747 tokens (54 tools)
- Tool definitions: C, 65/100
- Failing "description": 17 of 54 tools
- Failing "params_described": 43 of 54 tools
- Failing "annotations": 54 of 54 tools
- Warning "open_writes": prompts_create, prompts_update, prompts_delete, prompts_publish, runs_create
- Warning "destructive_unmarked": prompts_delete, runs_delete, datasets_delete, metrics_delete, tags_delete
- Method: https://inite.ai/en/atlas/mcp/report

## Your data

- Certifications stated: gdpr
- Training on your data: no (https://completionkit.com/legal/privacy)
- Read 2026-10-21

---

Source: INITE Atlas, https://inite.ai/en/atlas/completionkit.com. JSON: https://inite.ai/api/atlas/completionkit.com. MCP: https://inite.ai/api/mcp (atlas_lookup).