Metrx

{
  "schema_version": "1.0",
  "name": "Metrx",
  "description": "Metrx finds cheaper, better configurations for every AI workload, proves them against a randomized holdout, and switches on your say-so \u2014 as models and prices change. 23 MCP tools across 10 domains via npx; 38 across 14 domains on the hosted HTTP endpoint.",
  "url": "https://metrxbot.com",
  "provider": {
    "name": "Metrx AI",
    "url": "https://metrxbot.com"
  },
  "capabilities": {
    "mcp": {
      "server": "@metrxbot/mcp-server",
      "version": "0.2.5",
      "transport": {
        "stdio": "npx @metrxbot/mcp-server",
        "streamable_http": "https://metrxbot.com/api/mcp"
      },
      "tool_count": 23,
      "domains": [
        "cost-tracking",
        "optimization",
        "budget-governance",
        "alerts",
        "experiments",
        "cost-leak-detection",
        "attribution",
        "alert-configuration",
        "upgrade-justification",
        "roi-audit"
      ],
      "tools": {
        "metrx_get_cost_summary": "Get comprehensive cost summary for your AI agent fleet \u2014 total spend, call counts, error rates, agent breakdown, and optimization opportunities.",
        "metrx_list_agents": "List all registered agents with status, category, cost metrics, and health indicators.",
        "metrx_get_agent_detail": "Get detailed information about a specific agent including cost history, model usage, and performance metrics.",
        "metrx_get_optimization_recommendations": "Get AI-powered cost optimization recommendations \u2014 model switching, token guardrails, provider arbitrage, batch processing, and revenue intelligence.",
        "metrx_apply_optimization": "Stage a one-click optimization change for an agent under the authority you've granted (only for suggestions marked as one_click: true).",
        "metrx_route_model": "Get a model routing recommendation based on task complexity \u2014 recommends a cheaper model for simple tasks; routing executes only under a policy you enable.",
        "metrx_compare_models": "Compare LLM model pricing and capabilities across providers \u2014 works without usage data (Day 0 value).",
        "metrx_get_budget_status": "Get current budget status showing spending vs limits, warning/exceeded counts, and enforcement modes.",
        "metrx_set_budget": "Create or update a budget configuration with spending limits and enforcement modes (soft warn, hard cap, auto-pause).",
        "metrx_update_budget_mode": "Change enforcement mode or pause/resume an existing budget.",
        "metrx_get_alerts": "Get active alerts \u2014 cost spikes, error rate increases, budget warnings, and system health notifications.",
        "metrx_acknowledge_alert": "Mark alerts as read/acknowledged to clear notification state.",
        "metrx_get_failure_predictions": "Get predictive failure analysis \u2014 identifies agents likely to fail or exceed budgets before it happens.",
        "metrx_create_model_experiment": "Start an A/B test comparing two LLM models \u2014 tracks cost, latency, error rate, and quality until statistical significance.",
        "metrx_get_experiment_results": "Get current results and statistical significance of model experiments.",
        "metrx_stop_experiment": "Stop a running experiment with optional promotion of the winning model.",
        "metrx_run_cost_leak_scan": "Run comprehensive cost leak audit \u2014 identifies idle agents, model overprovisioning, missing caching, high error rates, context bloat, missing budgets, and arbitrage opportunities.",
        "metrx_configure_alert_threshold": "Set up cost or operational alert thresholds that trigger email, webhook, or auto-pause actions.",
        "metrx_attribute_task": "Link an agent task/event to a business outcome (revenue, cost saving, efficiency, quality) for ROI tracking.",
        "metrx_get_task_roi": "Calculate ROI for a specific agent \u2014 compares costs against attributed business outcomes.",
        "metrx_get_attribution_report": "Get attribution report showing business outcomes linked to agent actions with confidence scores.",
        "metrx_get_upgrade_justification": "Generate an ROI report recommending Verify (buyable per-verdict proof) and Platform (priced to your managed LLM spend) based on usage patterns and optimization potential.",
        "metrx_generate_roi_audit": "Generate comprehensive ROI audit report suitable for board reporting and compliance \u2014 per-agent cost/revenue breakdown with methodology."
      },
      "hosted": {
        "note": "The streamable-HTTP endpoint is built from the Metrx monorepo and is ahead of the published npm package. These tools are NOT available over stdio.",
        "url": "https://metrxbot.com/api/mcp",
        "version": "0.4.1",
        "tool_count": 38,
        "domains": [
          "cost-tracking",
          "optimization",
          "budget-governance",
          "alerts",
          "experiments",
          "cost-leak-detection",
          "attribution",
          "alert-configuration",
          "upgrade-justification",
          "roi-audit",
          "health-rankings",
          "upgrade-business-case",
          "verify-billing",
          "loop-metrics"
        ],
        "additional_tools": {
          "metrx_get_health_scores": "Get health score, grade, and data completeness for every agent \u2014 composite 0-100 scores blending cost efficiency, ROI, error rate, quality drift, failure risk, and latency.",
          "metrx_get_rankings": "Get agents sorted by health score with percentile ranking, fleet summary, and stale-score warnings \u2014 identifies top and bottom performers.",
          "metrx_get_upgrade_business_case": "Generate a personalized ROI business case for paying for Metrx \u2014 Verify to prove savings, then Platform (priced to your spend) \u2014 with projected savings and payback from real usage data.",
          "metrx_get_balance": "Check how many verifications your organization has remaining. Free to call \u2014 use before metrx_verify_switch to know whether a purchase is needed first.",
          "metrx_buy_verification": "Purchase one verification. Returns a Stripe checkout URL for a human (or payment-capable agent) to complete; this tool does not charge anything itself.",
          "metrx_verify_switch": "Start a purchased verification: would a candidate model hold quality vs the production model on your own traffic? Returns an async handle immediately and never blocks.",
          "metrx_get_signal_readiness": "Org-scoped signal-readiness diagnostics \u2014 event coverage, label lag, join rate, volume-vs-power, estimator eligibility, and blocked-on-signal proposals cross-linked to the failing dimension. Envelope-shaped response (measured-value envelopes; CIs never fabricated).",
          "metrx_get_dashboard_headline": "The three-number headline regime \u2014 measured spend, attributed (modeled) savings, and verified savings-with-CI, each with basis and provenance. Envelope-shaped response; requires the optimization_model_v3 flag.",
          "metrx_get_report_snapshot": "Fetch one frozen executive-report snapshot by public share token \u2014 stored values only, never recomputed, with snapshot_hash and methodology_version integrity stamps. Envelope-shaped response.",
          "metrx_get_verdicts": "Org-scoped current causal-verdict summaries (verdict, evidence mode/regime, pair counts, savings CI, expiry; :value rows flagged observational). Read-only twin route; evidence packs and gate actions stay dashboard-session-only.",
          "metrx_get_trials": "Org-scoped pre-registered trial plans: regime, primary metric, margin, frozen sample plan, status and concluding verdict id. Read-only.",
          "metrx_get_proposals": "Org-scoped current candidate proposals: action type/class, D-class, status with blocked_reason, estimated impact, taxonomy axes. Config payloads excluded. Read-only.",
          "metrx_get_actuations": "Org-scoped policy actuation records: action class, authority, rollback state and acknowledgement. Config contents excluded. Flag-gated (optimization_model_v3). Read-only.",
          "metrx_get_decision_queue": "The operator decision queue derived from real loop objects. Read-only: approve/snooze/acknowledge actions remain dashboard-session-only. Flag-gated (optimization_model_v3).",
          "metrx_kb_prior_lookup": "Per-cell (task_verb \u00d7 domain) summary of the caller org's OWN current causal verdicts \u2014 the knowledge-base read-side v0. Strictly org-local tenancy; honest zeros (unstamped_verdicts counts pre-taxonomy evidence). Allocation labels deliberately absent in v0. Read-only, envelope-shaped."
        }
      }
    },
    "rest_api": {
      "base_url": "https://metrxbot.com/api/v1",
      "auth": "Bearer token (API key from Settings \u2192 Security)",
      "docs": "https://docs.metrxbot.com/api-reference"
    },
    "otel": {
      "endpoint": "https://gateway.metrxbot.com/v1/traces",
      "protocol": "OTLP/HTTP"
    },
    "sdks": {
      "python": "metrxbot",
      "typescript": "@metrxbot/sdk"
    }
  },
  "authentication": {
    "type": "api_key",
    "header": "Authorization",
    "format": "Bearer <api_key>",
    "alternate_header": "X-Api-Key",
    "registration_url": "https://metrxbot.com/sign-up"
  },
  "pricing": {
    "free_tier": true,
    "url": "https://metrxbot.com/pricing"
  },
  "links": {
    "llms_txt": "https://metrxbot.com/llms.txt",
    "llms_full_txt": "https://metrxbot.com/llms-full.txt",
    "ucp_manifest": "https://metrxbot.com/.well-known/ucp.json",
    "benchmarks": "https://metrxbot.com/benchmarks",
    "agent_quickstart": "https://metrxbot.com/agent-quickstart",
    "npm": "https://www.npmjs.com/package/@metrxbot/mcp-server"
  }
}