CompletionKit

com-completionkit-evals declares 54 capabilities across 54 tools, with 2 critical issues.

97/100 exposureHigh exposure

Manifest recorded 2026-10-16 from tools/list on https://completionkit.com/mcp, published in the official MCP registry as com.completionkit/evals 1.0.0. The date changes only when the manifest does. Not verified by a maintainer.

Issues

These describe what this server's manifest declares. A product can provide controls that MCP has no way to express — token revocation, an admin console, audit logs — and this page cannot see them. The absence of a declaration is not evidence that a control is absent.

critical · no-revocation · All 54 tools

54 of 54 tools declare no way for the customer to switch them off once granted. This applies to every tool because MCP has no field in which to declare a revocation mechanism; the product may provide one outside MCP that a manifest cannot show.

critical · irreversible-uncapped · 9 of 54 tools

9 tools are classified as irreversible and declare no limit.

high · no-expiry · All 54 tools

54 of 54 tools declare no expiry. This applies to every tool because MCP has no field in which to declare an expiry; the product may provide one outside MCP that a manifest cannot show.

high · no-audit-trail · All 54 tools

54 of 54 tools declare no record of their actions that the customer can read. This applies to every tool because MCP has no field in which to declare an audit requirement; the product may provide one outside MCP that a manifest cannot show.

high · uncapped-write · 25 of 54 tools

25 tools can modify data and declare no limit.

low · uncapped-read · 20 of 54 tools

20 tools can read data and declare no limit.

Declared capabilities

How each tool was read. declared means the server's own annotation said so — self-reported and unverified. inferred is from the tool's name. assumed means nothing indicated either way.

Tool	Read as	Basis
`prompts_list`	`read`	inferred
`prompts_get`	`read`	inferred
`prompts_create`	`write`	inferred
`prompts_update`	`write`	inferred
`prompts_delete`	`delete`	inferred
`prompts_publish`	`send`	inferred
`prompts_suggest_improvement`	`write`	assumed
`runs_list`	`read`	inferred
`runs_get`	`read`	inferred
`runs_create`	`write`	inferred
`runs_update`	`write`	inferred
`runs_delete`	`delete`	inferred
`runs_generate`	`write`	assumed
`runs_regrade`	`write`	assumed
`runs_rerun`	`write`	inferred
`runs_retry_failures`	`write`	assumed
`responses_list`	`read`	inferred
`responses_get`	`read`	inferred
`datasets_list`	`read`	inferred
`datasets_get`	`read`	inferred
`datasets_create`	`write`	inferred
`datasets_update`	`write`	inferred
`datasets_delete`	`delete`	inferred
`datasets_create_from_url`	`write`	inferred
`metrics_list`	`read`	inferred
`metrics_get`	`read`	inferred
`metrics_create`	`write`	inferred
`metrics_update`	`write`	inferred
`metrics_delete`	`delete`	inferred
`metrics_suggest_variants`	`write`	assumed
`metric_groups_list`	`read`	inferred
`metric_groups_get`	`read`	inferred
`metric_groups_create`	`write`	inferred
`metric_groups_update`	`write`	inferred
`metric_groups_delete`	`delete`	inferred
`metric_versions_list`	`read`	inferred
`metric_versions_publish`	`send`	inferred
`metric_versions_dismiss`	`write`	inferred
`provider_credentials_list`	`read`	inferred
`provider_credentials_get`	`read`	inferred
`provider_credentials_create`	`write`	inferred
`provider_credentials_update`	`write`	inferred
`provider_credentials_delete`	`delete`	inferred
`tags_list`	`read`	inferred
`tags_get`	`read`	inferred
`tags_create`	`write`	inferred
`tags_update`	`write`	inferred
`tags_delete`	`delete`	inferred
`agreements_list`	`read`	inferred
`agreements_create`	`write`	inferred
`judges_replay`	`write`	inferred
`judges_compare`	`read`	inferred
`promptfoo_import`	`write`	assumed
`usage_get`	`read`	inferred

Record history

Earlier records for this server. A correction is added here, never written over the record it corrects.

Recorded	Exposure	Source
2026-10-11	97	tools/list on https://completionkit.com/mcp, published in the official MCP registry as com.completionkit/evals 1.0.0

Answering this question about your own agent? The self-serve stack report applies the same reading to your own stack or server and gives you a dated document you can hand to a customer's security review. It is generated in your browser, with nobody to wait for. Buying one does not change this page, its score or its place in the index — see Independence.

Maintain this server? If this page misreads your manifest, or your server has changed, submit its current tools/list response and it will be re-scored and dated. Corrections are published alongside the original, not in place of it.