CompletionKit
Prompt evals over MCP: run a prompt on your dataset, score each output 1-5 with an LLM judge.
Should I use this
Quality & Safety
Findings (1)
- LOWTool 'tags_update' description lacks action verbin tags_update
Based on automated analysis of tool definitions and protocol compliance.
Context Cost
This is the approximate number of tokens consumed each time the server's tools are loaded into a model's context. Higher counts reduce the attention available for other tasks.
Install
One-Click Install
Add this to your `claude_desktop_config.json` file:
{
"mcpServers": {
"evals": {
"url": "https://completionkit.com/mcp"
}
}
}Remote endpoints
https://completionkit.com/mcpstreamable-httpWhat it can do
Tool inventory
Tools (54)
🟢prompts_list
List all prompts
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢prompts_get(id)
Get a prompt by ID
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer",
"description": "Prompt ID"
}
},
"required": [
"id"
]
}🟡prompts_create(name, description, template, llm_model, tag_names)
Create a prompt
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"description": {
"type": "string"
},
"template": {
"type": "string"
},
"llm_model": {
"type": "string"
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name",
"template",
"llm_model"
]
}🟡prompts_update(id, name, description, template, llm_model, ...)
Update a prompt. If the prompt already has runs, this creates a new DRAFT version (current=false) rather than editing in place or publishing — promote it with prompts_publish — so an agent's edits don't go live without a gate. If it has no runs, it is updated in place.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
},
"description": {
"type": "string"
},
"template": {
"type": "string"
},
"llm_model": {
"type": "string"
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"id"
]
}🔴prompts_delete(id)
Delete a prompt
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡prompts_publish(id)
Publish a prompt version, making it the current version
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}⚪prompts_suggest_improvement(run_id)
Suggest an improved version of a prompt, grounded in a run's test results and judge feedback. Analyzes the run's responses, scores, and reviews, then returns reasoning plus a rewritten template (preserving {{variables}}) and persists it as a Suggestion. Requires a run that has a prompt (not a scoring-only run).
Input Schema
{
"type": "object",
"properties": {
"run_id": {
"type": "integer",
"description": "The run whose results ground the improvement."
}
},
"required": [
"run_id"
]
}🟢runs_list
List all runs
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢runs_get(id)
Get a run by ID, including "metric_averages": a per-metric breakdown with each metric's average score (or pass rate for checks), how many rows it graded, and how many scored low. Use this to find the metric dragging a prompt down without listing responses.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡runs_create(name, prompt_id, dataset_id, judge_model, temperature, ...)
Create a run. Omit prompt_id and provide output_column to score existing outputs by grading a pre-existing dataset column instead of generating new ones.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"prompt_id": {
"type": "integer"
},
"dataset_id": {
"type": "integer"
},
"judge_model": {
"type": "string"
},
"temperature": {
"type": "number",
"description": "Sampling temperature for generation, 0 to 1. Leave it unset, which is the default, and no temperature is sent at all, so the model applies its own. Most current frontier models refuse the parameter outright; set it only when you are targeting a model that honours it, such as anything served locally through Ollama. A refused value is re-sent without one and the run is flagged temperature_ignored."
},
"max_tokens": {
"type": "integer",
"description": "Cap on generated tokens per row. Leave unset to use the provider client's default, which is what silently truncates long outputs and makes the judge score malformed JSON. Set it to whatever the prompt uses in production so the eval matches."
},
"judge_temperature": {
"type": "number",
"description": "Sampling temperature for the judge, 0 to 1. Defaults to 0 so re-judging the same output gives the same score. Raise it only to measure judge variance on purpose; any value above 0 makes the run's scores irreproducible."
},
"output_column": {
"type": "string",
"description": "Dataset column to grade when prompt_id is omitted; defaults to \"actual_output\"."
},
"expected_column": {
"type": "string",
"description": "Dataset column holding each row's answer key / ground truth, graded by checks with compare_to \"expected\" and passed to the judge; defaults to \"expected_output\"."
},
"metric_ids": {
"type": "array",
"items": {
"type": "integer"
}
},
"metric_group_id": {
"type": "integer",
"description": "Attach the metrics belonging to this metric group (its current metric_ids). Ignored when metric_ids is also given."
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name"
]
}🟡runs_update(id, name, dataset_id, judge_model, temperature, ...)
Update a run
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
},
"dataset_id": {
"type": "integer"
},
"judge_model": {
"type": "string"
},
"temperature": {
"type": "number",
"description": "Sampling temperature for generation, 0 to 1. Leave it unset, which is the default, and no temperature is sent at all, so the model applies its own. Most current frontier models refuse the parameter outright; set it only when you are targeting a model that honours it, such as anything served locally through Ollama. A refused value is re-sent without one and the run is flagged temperature_ignored."
},
"max_tokens": {
"type": "integer",
"description": "Cap on generated tokens per row. Leave unset to use the provider client's default, which is what silently truncates long outputs and makes the judge score malformed JSON. Set it to whatever the prompt uses in production so the eval matches."
},
"judge_temperature": {
"type": "number",
"description": "Sampling temperature for the judge, 0 to 1. Defaults to 0 so re-judging the same output gives the same score. Raise it only to measure judge variance on purpose; any value above 0 makes the run's scores irreproducible."
},
"output_column": {
"type": "string"
},
"expected_column": {
"type": "string"
},
"metric_ids": {
"type": "array",
"items": {
"type": "integer"
}
},
"metric_group_id": {
"type": "integer",
"description": "Replace the run's metrics with those belonging to this metric group. Ignored when metric_ids is also given."
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"id"
]
}🔴runs_delete(id)
Delete a run
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}⚪runs_generate(id)
Start a run. Required for every run, including score-only runs (no prompt): generates responses with the prompt when there is one, otherwise copies the graded dataset column and grades it.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}⚪runs_regrade(id)
Re-grade a run's existing responses with its currently attached metrics, without regenerating. Use after attaching or editing metrics on an already-generated run.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡runs_rerun(id)
Create and start a fresh copy of a run with the same prompt, dataset, metrics, and settings. Use when the judge changed and you want a clean run instead of mixing versions.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}⚪runs_retry_failures(id, only)
Re-run only the failed responses of a run, optionally limited to specific response ids via "only".
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"only": {
"type": "array",
"items": {
"type": "integer"
}
}
},
"required": [
"id"
]
}🔴responses_list(run_id, limit, offset, status, min_score, ...)
List responses for a run, in row order. Returns {total, limit, offset, returned, responses}. Defaults to 50 rows because full payloads are large: use "fields" to drop the bodies, "min_score"/"max_score" to isolate low scorers, and sort "score_asc" to read the worst rows first. For per-metric averages of the whole run use runs_get instead of aggregating here.
Input Schema
{
"type": "object",
"properties": {
"run_id": {
"type": "integer"
},
"limit": {
"type": "integer",
"description": "Rows to return; defaults to 50, capped at 500."
},
"offset": {
"type": "integer",
"description": "Rows to skip before returning results."
},
"status": {
"type": "string",
"description": "Filter by row status: pending, retrying, succeeded or failed."
},
"min_score": {
"type": "number",
"description": "Only rows whose average judge score is at least this."
},
"max_score": {
"type": "number",
"description": "Only rows whose average judge score is at most this. Use with sort \"score_asc\" for failure-mode analysis."
},
"sort": {
"type": "string",
"enum": [
"id",
"score_asc",
"score_desc"
],
"description": "Row order; defaults to \"id\"."
},
"fields": {
"type": "array",
"items": {
"type": "string"
},
"description": "Only return these keys, keeping the payload small. Response keys: id, run_id, input_data, response_text, expected_output, created_at, score, reviewed, reviews, status, attempts, row_index, error. Prefix with \"reviews.\" to trim each review, e.g. [\"score\", \"reviews.metric_name\", \"reviews.ai_score\"]. id is always included."
}
},
"required": [
"run_id"
]
}🟢responses_get(run_id, id)
Get a specific response
Input Schema
{
"type": "object",
"properties": {
"run_id": {
"type": "integer"
},
"id": {
"type": "integer"
}
},
"required": [
"run_id",
"id"
]
}🟢datasets_list
List all datasets
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢datasets_get(id)
Get a dataset by ID
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡datasets_create(name, csv_data, tag_names)
Create a dataset with CSV data. First row is the header. Two column names are recognized specially: "expected_output" is each row's answer key (ground truth) given to the judge and to checks that compare against the row's expected value, and "actual_output" is a pre-made output to score in a prompt-less run. Both are overridable per run (expected_column / output_column). Every column is also available to the prompt as a variable.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"csv_data": {
"type": "string"
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name",
"csv_data"
]
}🟡datasets_update(id, name, csv_data, tag_names)
Update a dataset
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
},
"csv_data": {
"type": "string"
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"id"
]
}🔴datasets_delete(id)
Delete a dataset
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡datasets_create_from_url(name, url, tag_names)
Create a dataset by downloading CSV from a URL instead of inlining it. Use this for large datasets: pass a public http(s) URL and the server fetches the CSV directly, so the data never has to pass through the tool-call arguments. The URL is SSRF-checked and the download is capped at 10MB. First row is the header; the "expected_output" (answer key) and "actual_output" (pre-made output) columns are recognized specially, overridable per run.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"url": {
"type": "string",
"description": "Public http(s) URL of the CSV file to download."
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name",
"url"
]
}🟢metrics_list
List all metrics
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢metrics_get(id)
Get a metric by ID
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡metrics_create(name, instruction, metric_type, rubric_bands, check_config, ...)
Create a metric with evaluation criteria. For a deterministic check set metric_type:"check" and check_config. Per-kind required keys: value (contains/not_contains/equals), pattern (regex), json_path+expected (json_path_equals), min and/or max (length_bounds); valid_json takes no extra keys. target_path is required when target is json_path. For contains, not_contains, and equals, set compare_to:"expected" to grade against each row's own expected_output (ground truth) instead of a constant value (drop value); add expected_path to dig into the expected value when it is JSON.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"instruction": {
"type": "string"
},
"metric_type": {
"type": "string",
"enum": [
"llm_judge",
"check"
]
},
"rubric_bands": {
"type": "array",
"items": {
"type": "object",
"properties": {
"stars": {
"type": "integer"
},
"description": {
"type": "string"
}
}
}
},
"check_config": {
"type": "object",
"properties": {
"check_kind": {
"type": "string",
"enum": [
"contains",
"not_contains",
"equals",
"regex",
"valid_json",
"json_path_equals",
"length_bounds"
]
},
"target": {
"type": "string",
"enum": [
"response_text",
"input_data",
"json_path"
]
},
"target_path": {
"type": "string"
},
"value": {
"type": "string"
},
"pattern": {
"type": "string"
},
"json_path": {
"type": "string"
},
"expected": {},
"compare_to": {
"type": "string",
"enum": [
"constant",
"expected"
]
},
"expected_path": {
"type": "string"
},
"min": {
"type": "integer"
},
"max": {
"type": "integer"
},
"case_sensitive": {
"type": "boolean"
},
"multiline": {
"type": "boolean"
},
"trim": {
"type": "boolean"
}
}
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name"
]
}🟡metrics_update(id, name, instruction, metric_type, rubric_bands, ...)
Update a metric. For a deterministic check set metric_type:"check" and check_config. Per-kind required keys: value (contains/not_contains/equals), pattern (regex), json_path+expected (json_path_equals), min and/or max (length_bounds); valid_json takes no extra keys. target_path is required when target is json_path. For contains, not_contains, and equals, set compare_to:"expected" to grade against each row's own expected_output (ground truth) instead of a constant value (drop value); add expected_path to dig into the expected value when it is JSON.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
},
"instruction": {
"type": "string"
},
"metric_type": {
"type": "string",
"enum": [
"llm_judge",
"check"
]
},
"rubric_bands": {
"type": "array",
"items": {
"type": "object",
"properties": {
"stars": {
"type": "integer"
},
"description": {
"type": "string"
}
}
}
},
"check_config": {
"type": "object",
"properties": {
"check_kind": {
"type": "string",
"enum": [
"contains",
"not_contains",
"equals",
"regex",
"valid_json",
"json_path_equals",
"length_bounds"
]
},
"target": {
"type": "string",
"enum": [
"response_text",
"input_data",
"json_path"
]
},
"target_path": {
"type": "string"
},
"value": {
"type": "string"
},
"pattern": {
"type": "string"
},
"json_path": {
"type": "string"
},
"expected": {},
"compare_to": {
"type": "string",
"enum": [
"constant",
"expected"
]
},
"expected_path": {
"type": "string"
},
"min": {
"type": "integer"
},
"max": {
"type": "integer"
},
"case_sensitive": {
"type": "boolean"
},
"multiline": {
"type": "boolean"
},
"trim": {
"type": "boolean"
}
}
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"id"
]
}🔴metrics_delete(id)
Delete a metric
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}⚪metrics_suggest_variants(metric_id, count, model)
Ask the model to rewrite the metric's judge instruction in N variants targeted at the recent disagreements. Each variant is saved as a draft MetricVersion with source="suggestion". Returns the persisted drafts. Stripe-metering hooks fire via ActiveSupport::Notifications under completion_kit.judge_suggestion.generated.
Input Schema
{
"type": "object",
"properties": {
"metric_id": {
"type": "integer"
},
"count": {
"type": "integer",
"description": "How many variants to request (default 1, max 3). One focused rewrite beats five reworded copies."
},
"model": {
"type": "string",
"description": "Override the model used to generate variants. Defaults to the configured judge model or an available judging model."
}
},
"required": [
"metric_id"
]
}🟢metric_groups_list
List all metric groups
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢metric_groups_get(id)
Get a metric group by ID
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡metric_groups_create(name, description, metric_ids, tag_names)
Create a metric group
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"description": {
"type": "string"
},
"metric_ids": {
"type": "array",
"items": {
"type": "integer"
}
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"name"
]
}🟡metric_groups_update(id, name, description, metric_ids, tag_names)
Update a metric group
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
},
"description": {
"type": "string"
},
"metric_ids": {
"type": "array",
"items": {
"type": "integer"
}
},
"tag_names": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"id"
]
}🔴metric_groups_delete(id)
Delete a metric group
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟢metric_versions_list(metric_id)
List every MetricVersion (drafts + published) for a metric, newest first. Each row carries version_number, state, source, current flag, and timestamps.
Input Schema
{
"type": "object",
"properties": {
"metric_id": {
"type": "integer"
}
},
"required": [
"metric_id"
]
}🟡metric_versions_publish(metric_version_id)
Publish a MetricVersion as the live version of its metric. Works for both 'draft → published' and 'revert to an older published version → current'. Transactionally flips current, demotes peers, and writes the version's instruction + rubric_bands back onto the metric so the judge grades against it.
Input Schema
{
"type": "object",
"properties": {
"metric_version_id": {
"type": "integer"
}
},
"required": [
"metric_version_id"
]
}🔴metric_versions_dismiss(metric_version_id)
Destroy a draft MetricVersion (use for either source: 'edit' or source: 'suggestion'). Published versions are refused — to demote a published version, publish a different one as current instead.
Input Schema
{
"type": "object",
"properties": {
"metric_version_id": {
"type": "integer"
}
},
"required": [
"metric_version_id"
]
}🟢provider_credentials_list
List all provider credentials (API keys are not exposed)
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢provider_credentials_get(id)
Get a provider credential by ID (API key is not exposed)
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡provider_credentials_create(provider, api_key, api_endpoint, api_version)
Create a provider credential
Input Schema
{
"type": "object",
"properties": {
"provider": {
"type": "string",
"enum": [
"openai",
"anthropic",
"ollama",
"openrouter",
"azure_foundry"
]
},
"api_key": {
"type": "string"
},
"api_endpoint": {
"type": "string"
},
"api_version": {
"type": "string"
}
},
"required": [
"provider",
"api_key"
]
}🟡provider_credentials_update(id, provider, api_key, api_endpoint, api_version)
Update a provider credential
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"provider": {
"type": "string"
},
"api_key": {
"type": "string"
},
"api_endpoint": {
"type": "string"
},
"api_version": {
"type": "string"
}
},
"required": [
"id"
]
}🔴provider_credentials_delete(id)
Delete a provider credential
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟢tags_list
List all tags
Input Schema
{
"type": "object",
"properties": {},
"required": []
}🟢tags_get(id)
Get a tag by ID
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟡tags_create(name)
Create a tag. Color is auto-assigned.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
}
},
"required": [
"name"
]
}🟡tags_update(id, name)
Rename a tag.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"name": {
"type": "string"
}
},
"required": [
"id"
]
}🔴tags_delete(id)
Delete a tag. Removes the tag from every linked metric, prompt, run, and dataset.
Input Schema
{
"type": "object",
"properties": {
"id": {
"type": "integer"
}
},
"required": [
"id"
]
}🟢agreements_list(run_id, response_id, metric_id, created_by)
List agreements. Filter by run_id, response_id, metric_id, or created_by.
Input Schema
{
"type": "object",
"properties": {
"run_id": {
"type": "integer"
},
"response_id": {
"type": "integer"
},
"metric_id": {
"type": "integer"
},
"created_by": {
"type": "string"
}
},
"required": []
}🟡agreements_create(run_id, response_id, metric_id, verdict, corrected_score, ...)
Upsert an agreement for (run, response, metric, created_by). Verdict is one of agree, disagree, borderline. corrected_score (1..5) is required when verdict is 'disagree'.
Input Schema
{
"type": "object",
"properties": {
"run_id": {
"type": "integer"
},
"response_id": {
"type": "integer"
},
"metric_id": {
"type": "integer"
},
"verdict": {
"type": "string",
"enum": [
"agree",
"disagree",
"borderline"
]
},
"corrected_score": {
"type": "number"
},
"note": {
"type": "string"
},
"created_by": {
"type": "string"
}
},
"required": [
"run_id",
"response_id",
"metric_id",
"verdict"
]
}🟡judges_replay(name, metric_id, dataset_id, judge_model, output_column)
Create a scoring run for the current judge over a dataset's existing outputs (wraps runs_create with prompt_id omitted and output_column supplied). This only sets up the run; call runs_generate to actually re-judge the outputs so you can compare against human verdicts.
Input Schema
{
"type": "object",
"properties": {
"name": {
"type": "string"
},
"metric_id": {
"type": "integer"
},
"dataset_id": {
"type": "integer"
},
"judge_model": {
"type": "string"
},
"output_column": {
"type": "string",
"description": "Dataset column with the existing outputs to grade. Defaults to actual_output."
}
},
"required": [
"name",
"metric_id",
"dataset_id",
"judge_model"
]
}🟢judges_compare(metric_id, metric_version_a_id, metric_version_b_id)
Compare two versions of one metric's agreement stats side by side. Requires metric_id, metric_version_a_id, and metric_version_b_id (both versions must belong to that metric). Unavailable for check metrics.
Input Schema
{
"type": "object",
"properties": {
"metric_id": {
"type": "integer"
},
"metric_version_a_id": {
"type": "integer"
},
"metric_version_b_id": {
"type": "integer"
}
},
"required": [
"metric_id",
"metric_version_a_id",
"metric_version_b_id"
]
}🟢promptfoo_import(config)
Import a promptfooconfig.yaml. Creates a prompt, a dataset from the test vars, and metrics from the assert blocks (llm-rubric/g-eval become judge metrics; contains/equals/regex/is-json become deterministic check metrics). Returns a summary of what mapped and what was skipped and why; nothing is dropped silently.
Input Schema
{
"type": "object",
"properties": {
"config": {
"type": "string",
"description": "The full promptfooconfig.yaml contents."
}
},
"required": [
"config"
]
}🟢usage_get
Get this organization's plan usage and limits for the current billing period: runs and prompt fetches used, their limits, how many remain, and when the period resets. Call this to pre-check quota before starting runs. Runs are hard-blocked once the run limit is reached (with a small grace band), so a run over the limit will fail with run_limit_reached.
Input Schema
{
"type": "object",
"properties": {},
"required": []
}Community
Evidence