Skip to main content

AI Center

Use the Coralogix MCP server to work with AI Center directly from your AI agent: take inventory of your AI applications, read and configure evaluations and custom evaluation policies, link policies to applications, and manage per-model pricing.

These tools are available on MCP server v2 and inherit the permissions of your authenticated user (see AI Center permissions). The read tools need any one of AI-OVERVIEW:READ, AI-APP-CATALOG:READ, or AI-APP-EVALUATORS:READ. The evaluation write tools need AI-APP-EVALUATORS:MANAGE; the custom-evaluation and model-pricing tools require AI Center management access. Reading interaction content also requires SPANS.DATA-API#HIGH:READDATA.

Note

The write tools change live AI Center configuration. MCP clients prompt for per-call approval on the destructive ones (create, update, delete, link, unlink, set pricing) before they run.

Read tools

ToolDescription
ai_center_list_applicationsList the team's AI applications, including each app's guarded (guardrailsIntegrated) status.
ai_center_get_applicationGet one AI application by its UUID.
ai_center_list_evaluationsList configured evaluations (enabled state, target, config), optionally scoped to one app by application + subsystem.
ai_center_get_evaluationGet one configured evaluation by its UUID.
ai_center_get_coverageGet evaluation coverage across applications—a mapping of each evaluation type to the number of apps using it.
ai_center_list_custom_evaluationsList all custom evaluation policies for the team.
ai_center_list_custom_evaluations_for_applicationList custom evaluations linked to a specific application, by app UUID.
ai_center_get_model_pricingGet the team's custom per-model pricing overrides.

Write tools

ToolDescription
ai_center_create_evaluationCreate (enable) an evaluation on an AI application.
ai_center_update_evaluationUpdate a configured evaluation by UUID (enable/disable, change config or threshold). Partial patch.
ai_center_delete_evaluationDelete a configured evaluation by UUID, removing it from its application. Not reversible.
ai_center_create_custom_evaluationCreate a custom evaluation policy.
ai_center_update_custom_evaluationUpdate a custom evaluation policy by UUID. Partial patch.
ai_center_add_policy_to_applicationAttach a custom evaluation (policy) to an AI application.
ai_center_remove_policy_from_applicationDetach a custom evaluation (policy) from an application. Removes only the link; the policy itself is untouched and can be re-added.
ai_center_set_model_pricingSet the team's custom per-model pricing overrides. Team-wide; applies to new data only.
Note

Some deletions are intentionally not available through the MCP server: deleting a custom evaluation policy, deleting an AI application, and deleting model-pricing overrides. Use the Coralogix UI for those. To stop applying a policy to an app, unlink it with ai_center_remove_policy_from_application—the policy object survives.

Create and update parameters

ai_center_create_evaluation:

ParameterRequiredTypeDescription
applicationYesstringThe application to attach the evaluation to.
subsystemYesstringThe application's subsystem.
configYesobjectThe EvaluationConfig object—the $case variant plus its fields.
targetNoenumThe evaluated turn: prompt or response.
is_enabledNobooleanWhether the evaluation is enabled. Defaults to true.
thresholdNonumberScore threshold for the evaluation.

ai_center_create_custom_evaluation takes a body object with name, instructions, policyType, and optional description, violates, safe, examples, and applicationIds.

ai_center_update_evaluation and ai_center_update_custom_evaluation take the object's UUID plus an updates object—a partial patch, so include only the fields you want to change; omitted fields are left unchanged.

ai_center_set_model_pricing takes a prices object mapping each model name to { inputPricePerMillionTokens, outputPricePerMillionTokens, cacheReadPricePerMillionTokens, cacheWritePricePerMillionTokens }.

Example prompts

Take inventory

List my AI applications and tell me which ones aren't guarded.

Check evaluation coverage

Which evaluation types have the least coverage across my AI apps?

Configure an evaluation

Enable the hallucination evaluation on the "support-bot" app's production subsystem,
evaluating the response.
Attach my "no-legal-advice" custom policy to the support-bot application.

Next steps

Last updated on