Agent Eligibility Engine

npx add-mcp 'https://paynemuse.pythonanywhere.com/mcp'

Don’t let an agent guess whether another agent is qualified.

Payne Machine Foundry turns capabilities, evidence, policy, freshness, reliability, and fulfillment signals into a machine-readable CLEAR / HOLD / FAIL decision — so an orchestrator can decide, in one call, whether to hand an agent the job.

✓ Machine-readable ✓ Task-specific ✓ API-first ✓ Built for autonomous commerce

Agent supply is exploding. Selection is becoming the bottleneck.

Tool directories tell machines what exists. They do not reliably tell a machine: “Should I give this agent this job right now?” Every autonomous workflow eventually hits the same decision point — and today that decision is a guess, a hardcoded allowlist, or blind trust.

DISCOVER

→

QUALIFY the Foundry owns this

→

DECIDE the Foundry owns this

→

ASSIGN

→

VERIFY

→

PAY

The Foundry owns the QUALIFY / DECIDE layer: a deterministic, auditable eligibility check that sits between discovery and assignment.

Run an eligibility check.

Describe the agent, the job, and the evidence you have. The engine returns a decision, a score, and the exact reasons — no account, no payment, no guessing.

Decision

No evaluation yet.
Fill in the agent and the job, then run the check.

The scoring rules are public. Audit them.

Deterministic, versioned (rules v1.0.0), no hidden model. Start at 100, subtract for each failed check, clamp 0–100.

Check	Deduction
Missing required capability	−18 each
Payment not verified	−8
Delivery not verified	−10
Policy incompatible	−25
Source conflict	−25
Certification stale	−12
Failure rate > 10%	−20
Failure rate > 3%	−8
Historical jobs < 3	−8
Historical jobs < 10	−3

Decision bands. CLEAR ≥ 85 with no missing capabilities, no source conflict, and compatible policy. HOLD 60–84. FAIL < 60. Unknown evidence counts as unverified — the engine never assumes the best about evidence it hasn’t seen.

Built to be called by machines.

One endpoint. Structured JSON in, structured JSON out. No SDK required — curl, any HTTP client, or an MCP tool can call it.

POST/api/evaluate

Send the agent, the task, required capabilities, and whatever evidence you have. Get back decision, score, confidence, reasons[], and next_action — everything an orchestrator needs to decide whether to assign the job.

• No account required for the free tier
• Rate-limited per caller, abuse-protected
• Full contract at /api/openapi.json

curl -X POST https://foundry.example.com/api/evaluate \
  -H 'Content-Type: application/json' \
  -d '{
    "task": "Research a market question and return a cited current answer.",
    "required_capabilities": ["research", "web_search"],
    "agent": {
      "name": "agent-17",
      "capabilities": ["research", "web_search"],
      "evidence": {
        "payment_verified": true,
        "delivery_verified": true,
        "policy_compatible": true,
        "source_conflict": false,
        "certification_fresh": true,
        "failure_rate": 0.02,
        "historical_jobs": 27
      }
    }
  }'

# {"evaluation_id":"eval_...","decision":"CLEAR","score":100,
#  "confidence":"high","reasons":[],"next_action":"Eligible for
#  this task based on supplied evidence.", ...}

import requests

resp = requests.post(
    "https://foundry.example.com/api/evaluate",
    json={
        "task": "Research a market question and return a cited current answer.",
        "required_capabilities": ["research", "web_search"],
        "agent": {
            "name": "agent-17",
            "capabilities": ["research", "web_search"],
            "evidence": {
                "payment_verified": True,
                "delivery_verified": True,
                "policy_compatible": True,
                "source_conflict": False,
                "certification_fresh": True,
                "failure_rate": 0.02,
                "historical_jobs": 27,
            },
        },
    },
    timeout=10,
)
d = resp.json()
if d["decision"] == "CLEAR":
    assign_job("agent-17")   # proceed
elif d["decision"] == "HOLD":
    request_more_evidence()  # gather proof first
else:
    skip_agent()             # do not assign

const res = await fetch("https://foundry.example.com/api/evaluate", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    task: "Research a market question and return a cited current answer.",
    required_capabilities: ["research", "web_search"],
    agent: {
      name: "agent-17",
      capabilities: ["research", "web_search"],
      evidence: {
        payment_verified: true,
        delivery_verified: true,
        policy_compatible: true,
        source_conflict: false,
        certification_fresh: true,
        failure_rate: 0.02,
        historical_jobs: 27,
      },
    },
  }),
});
const d = await res.json();
// d.decision: "CLEAR" | "HOLD" | "FAIL"
// d.score, d.confidence, d.reasons[], d.next_action

Need this inside your agent stack?

We’re validating the first paid machine-eligibility use cases. Early users request access and help shape the API contract — the scoring rules, the evidence model, the receipt format.

Early access pricing — help shape the machine eligibility API.
Free evaluator now. Paid tiers (starter credit, developer monthly, usage-based) open only when the contract earns it. We’re looking for willingness to pay, not maximizing ARPU.