MCP Server

The WhiteBoxXAI MCP server is the MCP Governance Layer for AI assistants and agents. It gives any Model Context Protocol client (Claude Desktop, Claude Code, Cursor, VS Code, LangChain or Semantic Kernel agents, or your own harness) 197 tools over your WhiteBoxXAI data. With them, an assistant can register models, log predictions and LLM calls, check drift, explain predictions, run bias audits, triage alerts, read ISO/IEC 42001 readiness and the risk register, summarise governance review requests, and manage prompt versions.

Every tool calls the real WhiteBoxXAI API. None returns mocked or placeholder data. When something can't be computed, the tool says so.

Writing Python application code? Use the SDK instead. The MCP server is for assistants and agents that use WhiteBoxXAI as a tool.

Let your coding agent set this up

The whitebox-mcp-server skill teaches an agent to install and configure the server and pick the right tool. Load it with whitebox-governance-rules.

Two ways to connect

	Local (stdio)	Hosted (Streamable HTTP)
What runs where	whiteboxxai-mcp runs on your machine; your client starts it	Nothing to install; your client connects to https://api.whiteboxxai.com/mcp/
Credential	An API key in WHITEBOXXAI_MCP_API_KEY	An Authorization: Bearer header on every request: an API key or a personal sign-in token
Governance voting	Not available	Available with a personal sign-in token
Best for	Coding assistants and agents in your own environment	Clients that support remote MCP servers, and interactive voting

Install (local)

Requires Python 3.11 or later.

pip install whiteboxxai-mcp

Create an API key in Settings → API keys (administrators only; see API Keys), then check the server starts:

export WHITEBOXXAI_MCP_API_KEY="wbx_live_..."
whiteboxxai-mcp

Your MCP client normally starts the command itself; you don't keep it running in a terminal.

Configuration

Variable	Default	Purpose
WHITEBOXXAI_MCP_API_KEY	none (required)	Your API key (wbx_live_...)
WHITEBOXXAI_MCP_API_BASE_URL	https://api.whiteboxxai.com	Your API address. Business Cloud tenants use the address they were given.
WHITEBOXXAI_MCP_API_TIMEOUT	30	Request timeout, in seconds
WHITEBOXXAI_MCP_LOG_LEVEL		Log level, for example DEBUG

whiteboxxai-mcp --verbose (or WHITEBOXXAI_MCP_VERBOSE=1) logs request and response metadata, never credential values.

The local server authenticates only with an API key. If none is set, it exits with code 2 and whiteboxxai-mcp: no credentials configured.

Connect a client

Claude Desktop / Cursor / WindsurfClaude CodeVS CodeHosted (remote HTTP)LangChain / Semantic Kernel / custom

Add to the client's MCP config: claude_desktop_config.json for Claude Desktop, ~/.cursor/mcp.json (or .cursor/mcp.json in a project) for Cursor.

{
  "mcpServers": {
    "whiteboxxai": {
      "command": "whiteboxxai-mcp",
      "env": {
        "WHITEBOXXAI_MCP_API_BASE_URL": "https://api.whiteboxxai.com",
        "WHITEBOXXAI_MCP_API_KEY": "wbx_live_..."
      }
    }
  }
}

claude mcp add whiteboxxai \
  --env WHITEBOXXAI_MCP_API_KEY=wbx_live_... \
  -- whiteboxxai-mcp

Or add the same mcpServers block as Claude Desktop to a project's .mcp.json. Don't commit a real key; reference an environment variable such as ${WHITEBOXXAI_MCP_API_KEY} in the file.

VS Code uses a servers key in .vscode/mcp.json, and can prompt for the key instead of storing it:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "whitebox-key",
      "description": "WhiteBoxXAI API key",
      "password": true
    }
  ],
  "servers": {
    "whiteboxxai": {
      "type": "stdio",
      "command": "whiteboxxai-mcp",
      "env": {
        "WHITEBOXXAI_MCP_API_KEY": "${input:whitebox-key}"
      }
    }
  }
}

For clients that support remote MCP servers. With Claude Code:

claude mcp add --transport http whiteboxxai https://api.whiteboxxai.com/mcp/ \
  --header "Authorization: Bearer wbx_live_..."

In a JSON config that supports remote servers:

{
  "mcpServers": {
    "whiteboxxai": {
      "type": "http",
      "url": "https://api.whiteboxxai.com/mcp/",
      "headers": { "Authorization": "Bearer wbx_live_..." }
    }
  }
}

Any MCP client library can start whiteboxxai-mcp over stdio with the environment above, or connect to the hosted endpoint with an Authorization header. For LangChain, the langchain-mcp-adapters package loads the tools into an agent.

More copy-paste definitions are on Agent Config Definitions.

Hosted endpoint authentication

• Every request needs its own Authorization: Bearer <token> header: an API key, or a personal sign-in token. There's no fallback to environment variables, and a request without a valid header is refused with 401 before a session starts.
• A personal sign-in token expires after about 30 minutes and isn't refreshed by the server. Reconnect with a fresh token, or use an API key for long-running sessions.
• Activity sent with an API key is metered as Governance Events in the same way as SDK calls.

MCP Apps: interactive cards

In clients that support MCP Apps, some results render as interactive cards inside the conversation instead of raw JSON:

Card	Shown for
Explanation card	whitebox_explanations_generate, _get, _get_by_prediction
Alert triage card	whitebox_alerts_get_instance, whitebox_alerts_get_triage_context, whitebox_drift_get_report
Governance dossier card	whitebox_governance_list_review_requests, _get_review_request, _submit_decision
ISO/IEC 42001 status card	whitebox_compliance_iso42001_score, _gaps, _status
Model version diff card	whitebox_models_compare_versions, _version_changelog, _architecture_timeline

How the cards are built:

• Each card is a sandboxed ui://whitebox/... resource with a strict content security policy: it can't load anything from the network and holds no credentials. Buttons on a card ask your client to call a tool; they don't call WhiteBoxXAI directly.
• The card and the text result carry the same data and the same plain-language verdict. Clients without MCP Apps support get the full JSON result.
• Explanation cards replace common personal data patterns, such as email addresses and phone numbers, with [REDACTED:<TYPE>] before display.

Governance voting

The governance tools let an assistant prepare a review: list what's waiting, open a request's dossier (board, submitter, decisions, comments, vote tally and its evidence snapshot hash), and summarise it. Casting the vote stays with a person:

• whitebox_governance_submit_decision is an app-only tool. It's hidden from the model and only offered as a button on the governance dossier card, so an assistant can't vote on its own.
• A vote needs your own sign-in token on the hosted endpoint. An organization API key is refused, and so is the local stdio server, which only uses API keys.
• Every vote needs a current authenticator code, and two-factor authentication must be on.
• The vote carries the evidence snapshot hash of the dossier you saw. If the request changed since, it's refused with 409, and you review the current version first.

See Reviews & Decisions.

Example prompts

Ask the assistant	It calls
"What models do we have?"	whitebox_models_list
"Has credit_scoring drifted this month? Keep a record."	whitebox_drift_detect, then whitebox_drift_create_report
"Why was this prediction declined?"	whitebox_explanations_get_by_prediction or whitebox_explanations_generate
"What's firing right now, and why?"	whitebox_alerts_list_instances, whitebox_alerts_get_triage_context
"Where are we on ISO 42001?"	whitebox_compliance_iso42001_status
"What's waiting for my review?"	whitebox_governance_list_review_requests
"How did v3 differ from v2?"	whitebox_models_compare_versions
"Roll back the support-bot prompt to the last verified version."	whitebox_prompts_rollback

Tool reference

Tools are named whitebox_<domain>_<action>, across 18 domains. Required arguments are listed; see each tool's input schema in your client for optional ones.

Models

15 tools.

Tool	What it does	Required arguments
whitebox_models_architecture_timeline	Get the architecture changes timeline across all versions of a model.	model_id
whitebox_models_archive	Archive a model (soft delete; can be restored later).	model_id
whitebox_models_compare_versions	Compare two Git-backed model versions, returning architecture changes, parameter deltas, and breaking change flags.	version1_id, version2_id
whitebox_models_delete	Permanently delete a model.	model_id
whitebox_models_get	Get a registered model's full record by its model_id.	model_id
whitebox_models_latest	Get the most recently created version of a model by name.	model_name
whitebox_models_list	List registered models, optionally filtered by status/type/tags/search text.	—
whitebox_models_list_git_versions	List Git-backed version sources for a model, newest first.	model_id
whitebox_models_register	Register a new model in WhiteBoxXAI's model registry.	name, model_type, features, baseline_metrics
whitebox_models_restore	Restore a previously archived model to active status.	model_id
whitebox_models_update	Partially update a model's metadata.	model_id
whitebox_models_update_baseline	Refresh a model's baseline metrics and profile, for example after retraining.	model_id, baseline_metrics
whitebox_models_update_status	Update a model's status: ACTIVE, INACTIVE, DEPRECATED, or ARCHIVED.	model_id, new_status
whitebox_models_version_changelog	Generate a markdown changelog between two Git-backed model versions.	version1_id, version2_id
whitebox_models_versions	List all versions of a model by name, newest first.	model_name

Predictions

6 tools.

Tool	What it does	Required arguments
whitebox_predictions_get	Get a single logged prediction by its prediction_id.	prediction_id
whitebox_predictions_log	Log a single model prediction for monitoring.	model_id, input_data, output_data
whitebox_predictions_log_batch	Log up to 1000 predictions in a single batch call.	model_id, predictions
whitebox_predictions_query	Query logged predictions for a model, optionally within a time range.	model_id
whitebox_predictions_recent	Get the most recent predictions for a model (max 100).	model_id
whitebox_predictions_stats	Get prediction volume/latency statistics for a model.	model_id

Prompts

13 tools.

Tool	What it does	Required arguments
whitebox_prompts_advance_stage	Advance a prompt version through the lifecycle stages (design -> verify -> deploy -> retired).	template_id, version_id, target_stage
whitebox_prompts_compute_drift	Compute prompt drift analysis comparing production completions with baseline expectations.	template_id, version_id
whitebox_prompts_create_template	Create a new prompt template container for versioning.	name
whitebox_prompts_create_version	Create a new governed prompt version in 'design' lifecycle stage.	template_id, user_message_template
whitebox_prompts_deploy	Promote a prompt version to active production deployment.	template_id, version_id
whitebox_prompts_get_drift_report	Get a specific prompt drift report with metrics, divergence, and auditor verdict.	template_id, version_id, report_id
whitebox_prompts_get_incidents	Get security incidents and alerts associated with a prompt version.	template_id, version_id
whitebox_prompts_get_template	Get details of a specific prompt template by ID.	template_id
whitebox_prompts_get_version	Get details, lifecycle stage, and verification status of a prompt version.	template_id, version_id
whitebox_prompts_list_drift_reports	List historical prompt drift reports for a prompt version.	template_id, version_id
whitebox_prompts_list_templates	List all prompt governance templates for the authenticated tenant.	—
whitebox_prompts_list_versions	List all append-only governed versions for a prompt template.	template_id
whitebox_prompts_rollback	Emergency rollback to a target version or the most recently verified version.	template_id

Drift

5 tools.

Tool	What it does	Required arguments
whitebox_drift_create_report	Run drift analysis and persist the result as a drift report.	model_id
whitebox_drift_detect	Run an ad-hoc data drift check for a model (not persisted as a report).	model_id
whitebox_drift_get_report	Get a specific drift report, including per-feature statistics.	model_id, report_id
whitebox_drift_list_reports	List drift reports for a model, most recent first.	model_id
whitebox_drift_trend	Get a model's drift trend over time (1-90 days).	model_id

Fairness

6 tools.

Tool	What it does	Required arguments
whitebox_fairness_audit	Run a bias/fairness audit on model predictions across demographic groups.	model_id, sensitive_attributes, y_true, y_pred, group_data
whitebox_fairness_bias_history	Get a model's historical bias/fairness trend (1-365 days).	model_id
whitebox_fairness_get_audit	Get full bias audit results by audit_id.	audit_id
whitebox_fairness_latest_audit	Get the most recent bias audit for a model.	model_id
whitebox_fairness_list_audits	List bias audit summaries, optionally filtered to one model.	—
whitebox_fairness_metric_history	Get history for one specific fairness metric on a model.	model_id, metric_type

Explanations

13 tools.

Tool	What it does	Required arguments
whitebox_explanations_compare	Compare 2-20 explanations against each other.	explanation_ids
whitebox_explanations_delete_config	Delete a model's explanation configuration.	model_id
whitebox_explanations_generate	Generate a SHAP/LIME explanation and block until it's computed.	model_id, instance
whitebox_explanations_generate_async	Start real SHAP/LIME explanation generation for a model instance.	model_id, instance
whitebox_explanations_generate_bulk	Start explanation generation for multiple instances at once (non-blocking).	model_id, instances
whitebox_explanations_get	Get an explanation by ID, including its status and (once ready) results.	explanation_id
whitebox_explanations_get_by_prediction	Get the explanation associated with a prediction, if one exists.	prediction_id
whitebox_explanations_get_config	Get a model's explanation configuration.	model_id
whitebox_explanations_list_by_model	List explanations for a model, optionally filtered by method or status.	model_id
whitebox_explanations_set_config	Create or update a model's explanation configuration (upsert).	model_id
whitebox_explanations_stats	Get aggregated explanation statistics for a model (counts by method/status, average computation time).	model_id
whitebox_explanations_visualize	Get visualization data for a single explanation.	explanation_id, plot_type
whitebox_explanations_visualize_multi	Get visualization data spanning multiple explanations.	explanation_ids, plot_type

Risk register

3 tools.

Tool	What it does	Required arguments
whitebox_risk_register_get	Get a single AI risk register entry by ID.	entry_id
whitebox_risk_register_list	List AI risk register entries, optionally filtered.	—
whitebox_risk_register_portfolio	Get the likelihood x impact heat map and top open risks across the organization's risk register.	—

Governance

5 tools.

Tool	What it does	Required arguments
whitebox_governance_get_review_request	Get a specific governance review request dossier (board, submitter, decisions, comments, vote tally, and SHA-256 evidence snapshot hash).	request_id
whitebox_governance_list_boards	List governance review boards for the organization.	—
whitebox_governance_list_review_requests	List governance review requests, optionally filtered.	—
whitebox_governance_raci_grid	Get the RACI grid: review requests and their Responsible / Accountable / Consulted / Informed assignments (ISO 42001 Clause 5.3, Annex A.3), optionally scoped to one board.	—
whitebox_governance_submit_decision	Record a board member's vote on a governance review request.	request_id, decision, evidence_snapshot_hash, totp_code

Compliance

5 tools.

Tool	What it does	Required arguments
whitebox_compliance_get_framework_controls	Get the full control catalog for one compliance framework.	framework
whitebox_compliance_iso42001_gaps	Get the organization's open ISO/IEC 42001 compliance gaps.	—
whitebox_compliance_iso42001_score	Get the organization's ISO/IEC 42001 readiness score and clause breakdown.	—
whitebox_compliance_iso42001_status	Get a composite ISO/IEC 42001 readiness scorecard, open gap matrix, and linked AI risk register entries in a single call for the ui://whitebox/iso42001-status-card widget.	—
whitebox_compliance_list_frameworks	List all supported compliance frameworks in the Control Catalog (ISO/IEC 42001, GDPR, CCPA, EU AI Act, NIST AI RMF) with control counts.	—

LLM explainability

13 tools.

Tool	What it does	Required arguments
whitebox_llm_xai_attention	Analyze transformer attention patterns for a prompt/completion pair.	prompt
whitebox_llm_xai_batch_analyze	Run multiple explanation analyses on a single LLM log in one call (more efficient than calling each analysis tool separately).	llm_log_id
whitebox_llm_xai_counterfactuals	Generate minimal prompt edits that change the model's output, useful for understanding decision boundaries.	prompt, original_output
whitebox_llm_xai_debug_prompt	Assess prompt quality across multiple dimensions and get improvement recommendations.	prompt
whitebox_llm_xai_get_explanation	Get a specific attention/token-importance/sensitivity/counterfactual explanation by ID.	explanation_id
whitebox_llm_xai_get_prompt_analysis	Get a specific prompt debug analysis by ID.	analysis_id
whitebox_llm_xai_list_explanations	List explanations, optionally filtered by LLM log, type (attention_analysis/token_importance/sensitivity_analysis/ counterfactual_generation), or model.	—
whitebox_llm_xai_list_prompt_analyses	List prompt analyses, optionally filtered by similar prompt text, model, or minimum quality score (0-1).	—
whitebox_llm_xai_prompt_sensitivity	Test prompt robustness by perturbing it and measuring output stability.	prompt, baseline_output
whitebox_llm_xai_stats	Get aggregated LLM-XAI usage statistics (counts by explanation type, average computation time, prompt quality).	—
whitebox_llm_xai_token_importance	Calculate token-level importance scores for a prompt/completion pair.	prompt, completion
whitebox_llm_xai_visualize_attention	Get attention-heatmap visualization data for an explanation (for rendering with Plotly/D3.js-style tools).	explanation_id
whitebox_llm_xai_visualize_token_importance	Get token-importance bar-chart visualization data for an explanation.	explanation_id

LLM monitoring

16 tools.

Tool	What it does	Required arguments
whitebox_llm_monitoring_cleanup_logs	Delete LLM logs older than days.	—
whitebox_llm_monitoring_cost_breakdown	Get detailed cost breakdown (platform-wide aggregate).	—
whitebox_llm_monitoring_cost_threshold_alert	Check whether costs over the recent period exceed a threshold.	—
whitebox_llm_monitoring_error_rate_alert	Check whether the error rate over the recent period exceeds a threshold.	—
whitebox_llm_monitoring_get_log	Get a single LLM log by ID.	log_id
whitebox_llm_monitoring_latency_threshold_alert	Check whether average latency over the recent period exceeds a threshold.	—
whitebox_llm_monitoring_log_call	Log a single LLM call with real token/cost/latency data.	provider, model_name, prompt, latency_ms
whitebox_llm_monitoring_log_calls_batch	Log multiple LLM calls in one request (max 1000).	logs
whitebox_llm_monitoring_performance	Get LLM latency/throughput performance metrics.	—
whitebox_llm_monitoring_query_logs	Query LLM logs with filters and pagination.	—
whitebox_llm_monitoring_recent	Get recent LLM logs (requires a model_id owned by the caller's org).	—
whitebox_llm_monitoring_session_logs	Get all LLM logs for a specific session.	session_id
whitebox_llm_monitoring_trends_costs	Get cost trend over time (platform-wide; granularity: hour/day/week).	—
whitebox_llm_monitoring_trends_tokens	Get token usage trend over time (granularity: hour/day/week).	—
whitebox_llm_monitoring_usage_stats	Get aggregated LLM usage statistics (tokens, cost, latency, by_model).	—
whitebox_llm_monitoring_usage_stats_detailed	Get detailed, grouped usage statistics (granularity: hour/day/week/month).	start_time, end_time

RAG

10 tools.

Tool	What it does	Required arguments
whitebox_rag_create_evaluation	Create a RAG evaluation, aggregating already-logged retrievals for each query.	name, queries
whitebox_rag_get_evaluation	Get a specific RAG evaluation by ID.	evaluation_id
whitebox_rag_get_retrieval	Get a specific retrieval log by ID.	log_id
whitebox_rag_list_evaluations	List RAG evaluations with optional filters.	—
whitebox_rag_list_retrievals	Query retrieval logs for a model.	model_id
whitebox_rag_log_retrieval	Log a retrieval operation, with real quality metrics computed server-side when ground_truth_ids is supplied.	query, results, top_k, retrieval_method
whitebox_rag_metrics_precision	Get Precision@K metrics for the given K values (default [1, 3, 5, 10]).	—
whitebox_rag_metrics_relevance	Get average relevance metrics (context/answer/faithfulness).	—
whitebox_rag_stats	Get aggregated RAG statistics (precision/recall/MRR/NDCG/relevance/latency).	—
whitebox_rag_trends	Get RAG quality metrics trend over time (granularity: hour/day/week).	—

Safety

11 tools.

Tool	What it does	Required arguments
whitebox_safety_analyze	Analyze content for toxicity, PII, and harmful content.	content
whitebox_safety_analyze_batch	Analyze multiple content items in one request (max 100).	contents
whitebox_safety_create_threshold	Create a safety threshold configuration controlling when content is flagged as unsafe.	name
whitebox_safety_delete_threshold	Delete a safety threshold configuration.	threshold_id
whitebox_safety_get_score	Get a specific safety score by ID.	score_id
whitebox_safety_get_scores	Query safety analysis scores with filters.	—
whitebox_safety_get_threshold	Get a specific safety threshold by ID.	threshold_id
whitebox_safety_list_thresholds	List safety thresholds for the caller's organization.	—
whitebox_safety_stats	Get aggregated safety statistics (per-category toxicity averages, flagged categories, PII types).	—
whitebox_safety_trends	Get content-safety trend over time (granularity: hour/day/week).	—
whitebox_safety_update_threshold	Update a safety threshold configuration.	threshold_id

Alerts

13 tools.

Tool	What it does	Required arguments
whitebox_alerts_acknowledge	Acknowledge an active alert instance.	alert_id, user_id
whitebox_alerts_create_rule	Create an alert rule with threshold or anomaly-based conditions.	name, alert_type, severity, conditions
whitebox_alerts_delete_rule	Delete an alert rule.	rule_id
whitebox_alerts_evaluate_rule	Test whether an alert rule would trigger given metric values, without creating a real alert instance.	rule_id
whitebox_alerts_get_instance	Get a specific triggered alert instance by ID.	alert_id
whitebox_alerts_get_rule	Get a specific alert rule by ID.	rule_id
whitebox_alerts_get_triage_context	Get a triggered alert instance together with its model's latest drift reports and a plain-language triage verdict.	alert_id
whitebox_alerts_list_instances	List triggered alert instances with optional filters.	—
whitebox_alerts_list_rules	List alert rules for the caller's organization.	—
whitebox_alerts_resolve	Resolve an alert instance.	alert_id, user_id
whitebox_alerts_snooze	Snooze an active alert instance for snooze_minutes (max 1 week).	alert_id, snooze_minutes
whitebox_alerts_statistics	Get alert statistics for the organization (counts, severity distribution, mean time to acknowledge/resolve) over the last hours.	—
whitebox_alerts_update_rule	Update an alert rule.	rule_id

Alert intelligence

20 tools.

Tool	What it does	Required arguments
whitebox_alert_intelligence_adjust_priority	Manually override a calculated priority.	alert_id, new_priority_score, new_priority_level, adjustment_reason
whitebox_alert_intelligence_analyze_root_cause	Analyze potential root causes (data drift, model degradation, feature anomalies, system issues) for an alert.	alert_id
whitebox_alert_intelligence_apply_fatigue_reduction	Apply automated throttling/grouping/snoozing to reduce alert noise.	—
whitebox_alert_intelligence_calculate_priority	Compute a weighted priority score (0-100) and level (critical/high/medium/low) for an alert.	alert_id
whitebox_alert_intelligence_deactivate_pattern	Mark a pattern inactive (resolved or no longer relevant).	pattern_id
whitebox_alert_intelligence_detect_correlations	Detect correlations for an alert (temporal, model, feature, and pattern-based relationships to other alerts).	alert_id
whitebox_alert_intelligence_detect_patterns	Detect recurring temporal/frequency/correlation/anomaly patterns in alerts.	—
whitebox_alert_intelligence_detect_seasonality	Detect hour-of-day seasonality in alert volume.	—
whitebox_alert_intelligence_enhanced_trend_analysis	Combine alert-volume time series, an hour-of-day seasonality estimate, a linear trend prediction with an R²-based confidence score, and z-score anomaly detection, all computed from the organization's actual alert history.	—
whitebox_alert_intelligence_forecast_volume	Forecast future alert volume with confidence intervals.	—
whitebox_alert_intelligence_get_correlation_clusters	Get clusters of highly correlated alerts: useful for spotting widespread issues or cascading failures.	—
whitebox_alert_intelligence_get_pattern	Get a specific alert pattern's details.	pattern_id
whitebox_alert_intelligence_get_priority_distribution	Get a breakdown of alerts by priority level.	—
whitebox_alert_intelligence_get_priority_queue	Get the prioritized alert queue for triage and assignment.	—
whitebox_alert_intelligence_get_similar_incidents	Find past alerts sharing the same root-cause type: useful for learning from historical resolutions.	cause_type
whitebox_alert_intelligence_get_throttle_recommendations	Get recommended alert-throttling rules to reduce fatigue without missing critical issues.	—
whitebox_alert_intelligence_ml_grouping	Cluster alerts via ML (kmeans/dbscan/hierarchical).	—
whitebox_alert_intelligence_predict_trends	Forecast alert volume/severity/type for time_horizon (next_hour/next_day/next_week).	—
whitebox_alert_intelligence_temporal_grouping	Group alerts occurring in the same time window with similar characteristics.	—
whitebox_alert_intelligence_verify_root_cause	Mark a previously computed root cause as verified or rejected.	root_cause_id, verified

Alert management

13 tools.

Tool	What it does	Required arguments
whitebox_alert_management_acknowledge	Acknowledge an alert (hits /alert-management/alerts/{id}/acknowledge; distinct from whitebox_alerts_acknowledge).	alert_id
whitebox_alert_management_create_comment	Add a comment to an alert (comment_type: note/action/resolution).	alert_id, comment
whitebox_alert_management_detect_fatigue	Detect alert-fatigue indicators and get recommendations for reducing alert noise.	—
whitebox_alert_management_get_analytics	Get comprehensive alert analytics (counts, MTTA/MTTR, distributions, escalation metrics) for the caller's organization.	—
whitebox_alert_management_get_comments	List all comments on an alert.	alert_id
whitebox_alert_management_get_grouping	Group active alerts by a dimension: "type", "severity", "model", or "rule": useful for spotting hotspots.	group_by
whitebox_alert_management_get_history	Get the full action timeline (triggered/acknowledged/snoozed/ resolved/escalated/commented) for an alert.	alert_id
whitebox_alert_management_get_trend	Get alert volume as a time series between start and end (ISO 8601 timestamps), bucketed by interval (hour/day/week).	start, end
whitebox_alert_management_get_velocity	Get alert generation rate trends (hour-over-hour, day-over-day, rolling averages) for the caller's organization.	—
whitebox_alert_management_process_pending_escalations	Manually trigger processing of pending escalations (normally a scheduled background task).	—
whitebox_alert_management_resolve	Resolve an alert (hits /alert-management/alerts/{id}/resolve; distinct from whitebox_alerts_resolve).	alert_id
whitebox_alert_management_snooze	Snooze an alert for snooze_minutes (max 1 week; hits /alert-management/alerts/{id}/snooze, distinct from whitebox_alerts_snooze).	alert_id, snooze_minutes
whitebox_alert_management_unsnooze_expired_alerts	Manually trigger reactivation of alerts whose snooze period has expired (normally a scheduled background task).	—

Agent workflows

18 tools.

Tool	What it does	Required arguments
whitebox_agent_workflows_analytics	Get workflow analytics (tokens, costs, execution counts, durations).	workflow_id
whitebox_agent_workflows_bottlenecks	Identify workflow bottlenecks: slowest agents, slowest tasks, failing agents.	workflow_id
whitebox_agent_workflows_complete	Mark a workflow completed/failed/cancelled and optionally trigger async analytics (metrics, cost, bottlenecks, timeline).	workflow_id, status
whitebox_agent_workflows_cost_breakdown	Get per-agent cost breakdown (tokens, cost, execution count, avg duration).	workflow_id
whitebox_agent_workflows_create_and_start	Create a new multi-agent workflow.	name, framework
whitebox_agent_workflows_create_execution	Log that an agent started executing.	workflow_id, agent_id
whitebox_agent_workflows_create_task	Create a workflow task.	workflow_id, task_name
whitebox_agent_workflows_delete	Delete a workflow and all related agents/executions/tasks/interactions.	workflow_id
whitebox_agent_workflows_get	Get a workflow by ID.	workflow_id
whitebox_agent_workflows_list	List workflows for the organization.	—
whitebox_agent_workflows_list_agents	List all agents registered in a workflow.	workflow_id
whitebox_agent_workflows_list_interactions	List all agent-to-agent interactions in a workflow, chronologically.	workflow_id
whitebox_agent_workflows_list_tasks	List all tasks in a workflow, with status, assignments, and results.	workflow_id
whitebox_agent_workflows_log_interaction	Log agent-to-agent communication.	workflow_id, interaction_type
whitebox_agent_workflows_register_agent	Register an agent in a workflow.	workflow_id, name
whitebox_agent_workflows_start	Mark a workflow as running.	workflow_id
whitebox_agent_workflows_timeline	Get a chronological timeline of workflow events (start/end, agent executions, task updates, interactions).	workflow_id
whitebox_agent_workflows_update_task_status	Update a task's status.	task_id, status

Metrics

12 tools.

Tool	What it does	Required arguments
whitebox_metrics_aggregate	Get metrics aggregated over a period, such as daily or weekly.	model_id, period
whitebox_metrics_calculate_classification	Calculate classification metrics (accuracy/precision/recall/F1/etc.) from predictions, without persisting anything.	y_true, y_pred
whitebox_metrics_calculate_regression	Calculate regression metrics (MAE/MSE/RMSE/R2/etc.) from predictions, without persisting anything.	y_true, y_pred
whitebox_metrics_create	Create a new metric record for a model.	model_id, metric_type, metric_name, metric_value
whitebox_metrics_create_batch	Create multiple metrics at once (max 100).	model_id, metrics
whitebox_metrics_delete	Delete all metrics for a model.	model_id
whitebox_metrics_get_model_metrics	Get metrics recorded for a model, with optional type/date filters.	model_id
whitebox_metrics_latest	Get the most recent metric of a specific type for a model.	model_id, metric_type
whitebox_metrics_rolling	Get rolling-window statistics for a metric over window_days.	model_id, metric_type
whitebox_metrics_summary	Get a summary of all metrics for a model over the last days.	model_id
whitebox_metrics_timeseries	Get time series data for a metric over a required date range.	model_id, metric_type, start_date, end_date
whitebox_metrics_trend	Get trend statistics for a metric over lookback_days.	model_id, metric_type

Current limitations

• No report generation tools. Generate reports with the SDK or the dashboard. Over MCP, an assistant can create and read drift reports and read the evidence reports draw on.
• Not every administrative endpoint is a tool. Escalation-policy management and bulk alert actions are left out on purpose. If a tool doesn't exist, the assistant should say so rather than simulate it.
• Workflow analytics can be slow. whitebox_agent_workflows_cost_breakdown, _bottlenecks and _timeline can take up to about 30 seconds.
• Hosted sessions with a personal sign-in token expire after about 30 minutes; reconnect with a fresh token, or use an API key where you don't need to vote.

Troubleshooting

whiteboxxai-mcp: no credentials configured (exit code 2). Set WHITEBOXXAI_MCP_API_KEY in the environment the client starts the server with, not only in your shell.

401 Unauthorized. The key is mistyped, revoked or expired, or (on the hosted endpoint) the Authorization header is missing or the sign-in token has expired.

403 on a vote. Votes need your own sign-in token, two-factor authentication and an authenticator code. An API key can't vote.

404 on an id. The record isn't in your organization. Check the id rather than retrying with other credentials.

429. Either the rate limit (wait for Retry-After) or, on the Free plan, the month's Governance Events are used up.

Calls hang, then time out. WHITEBOXXAI_MCP_API_BASE_URL isn't reachable from where the server runs. Check the address and WHITEBOXXAI_MCP_API_TIMEOUT.

Security notes

• Use one API key per client or agent, so you can revoke one without breaking the others.
• Credential values are never logged; verbose mode logs request metadata only.
• The server doesn't write model weights or training data anywhere. It only calls the WhiteBoxXAI API.
• An API key acts with the permissions of the administrator who created it. See API Keys.

Related

• Agent Skills and AI Prompts
• API Keys
• Python SDK and the REST API reference