@paybond/kit 0.12.2 → 0.12.4

package/dist/mcp-server.js

@@ -15,12 +15,51 @@ import { verifyAgentReceiptV1FromJSON } from "./agent-receipt.js";
 import { agentReceiptResourceTemplateDefinition, agentReceiptResourceUri, MCP_AGENT_RECEIPT_RESOURCE_MIME_TYPE, parseAgentReceiptResourceUri, } from "./mcp-receipt-resource.js";
 import { DEFAULT_PAYBOND_GATEWAY_BASE_URL, GatewayFraudClient, GatewaySignalClient, } from "./index.js";
 const SERVER_NAME = "Paybond MCP";
-const SERVER_VERSION = "0.12.2";
+const SERVER_VERSION = "0.12.4";
 const MCP_PROTOCOL_VERSION = "2025-11-25";
 const DEFAULT_PRINCIPAL_PATH = "/v1/auth/principal";
 const DEFAULT_RECOGNITION_VERIFIER_ID = "paybond-gateway";
 const agentRecognitionProofHeader = "x-paybond-agent-recognition-proof";
 const DEFAULT_ENV_FILE = ".env.local";
+const AUTHORIZE_SPEND_OUTPUT_PROPERTIES = {
+    allow: {
+        type: "boolean",
+        description: "Whether the requested operation is allowed.",
+    },
+    tenant: {
+        type: "string",
+        description: "Tenant echoed by the gateway.",
+    },
+    intent_id: {
+        type: "string",
+        description: "Verified Harbor intent UUID.",
+    },
+    audit_id: {
+        type: "string",
+        description: "Gateway audit identifier when available.",
+    },
+    remaining_cents: {
+        type: "integer",
+        description: "Remaining spend budget in cents for the evaluated scope, when available.",
+    },
+    reason_codes: {
+        type: "array",
+        items: { type: "string" },
+        description: "Stable spend-policy reason codes from the authorization decision.",
+    },
+    message: {
+        type: "string",
+        description: "Human-readable decision message when present.",
+    },
+    decision_id: {
+        type: "string",
+        description: "Persisted spend decision identifier when authorization creates one.",
+    },
+    approval_request_id: {
+        type: "string",
+        description: "Approval request identifier when human approval is required.",
+    },
+};
 const TOOL_SELECTION_METADATA = {
     paybond_get_principal: {
         title: "Get Paybond Principal",
@@ -40,22 +79,7 @@ const TOOL_SELECTION_METADATA = {
             "Do not use this to create, fund, or modify intents; use paybond_authorize_agent_spend as the clearer gate before side-effecting agent tools.",
         annotations: additiveMutationToolAnnotations("Verify Paybond Capability"),
         outputSchema: outputObjectSchema({
-            allow: {
-                type: "boolean",
-                description: "Whether the requested operation is allowed.",
-            },
-            tenant: {
-                type: "string",
-                description: "Tenant echoed by the gateway.",
-            },
-            intent_id: {
-                type: "string",
-                description: "Verified Harbor intent UUID.",
-            },
-            audit_id: {
-                type: "string",
-                description: "Gateway audit identifier when available.",
-            },
+            ...AUTHORIZE_SPEND_OUTPUT_PROPERTIES,
         }, ["tenant", "intent_id"]),
     },
     paybond_authorize_agent_spend: {
@@ -64,23 +88,58 @@ const TOOL_SELECTION_METADATA = {
             "Do not use this for creating, funding, or changing intents; call paybond_create_spend_intent or paybond_fund_intent first when no funded capability token exists.",
         annotations: additiveMutationToolAnnotations("Authorize Agent Spend"),
         outputSchema: outputObjectSchema({
-            allow: {
-                type: "boolean",
-                description: "Whether the requested operation is allowed.",
+            ...AUTHORIZE_SPEND_OUTPUT_PROPERTIES,
+        }, ["tenant", "intent_id"]),
+    },
+    paybond_get_budget_remaining: {
+        title: "Get Budget Remaining",
+        description: "Use this when you need a read-only dry-run of remaining spend budget for a tenant-bound intent before authorizing a paid tool. " +
+            "Do not use this to authorize spend or create decisions; call paybond_authorize_agent_spend when you are ready to gate a side-effecting tool.",
+        annotations: readOnlyToolAnnotations("Get Budget Remaining"),
+        outputSchema: outputObjectSchema({
+            remaining_cents: {
+                type: "integer",
+                description: "Remaining spend budget in cents for the evaluated scope, when available.",
             },
-            tenant: {
-                type: "string",
-                description: "Tenant echoed by the gateway.",
+            spend_scope: {
+                type: "object",
+                description: "Spend scope used for the budget evaluation (scope_type and scope_key).",
+                additionalProperties: true,
             },
-            intent_id: {
+            policy_version: {
+                type: "integer",
+                description: "Active spend-control policy version when a policy is configured.",
+            },
+        }, []),
+    },
+    paybond_explain_policy: {
+        title: "Explain Spend Policy",
+        description: "Use this when you need a read-only explanation of whether a proposed spend would allow, require approval, or deny under the tenant spend-control policy. " +
+            "Do not use this to authorize spend or create approval requests; call paybond_authorize_agent_spend to persist a decision.",
+        annotations: readOnlyToolAnnotations("Explain Spend Policy"),
+        outputSchema: outputObjectSchema({
+            outcome: {
                 type: "string",
-                description: "Verified Harbor intent UUID.",
+                description: "Normalized policy outcome: allow, approval_required, or deny.",
             },
-            audit_id: {
+            reason_codes: {
+                type: "array",
+                items: { type: "string" },
+                description: "Stable policy reason codes from the dry-run evaluation.",
+            },
+            explanation: {
                 type: "string",
-                description: "Gateway audit identifier when available.",
+                description: "Human-readable explanation derived from reason codes.",
             },
-        }, ["tenant", "intent_id"]),
+            remaining_cents: {
+                type: "integer",
+                description: "Remaining spend budget in cents for the evaluated scope, when available.",
+            },
+            approval_threshold_exceeded: {
+                type: "boolean",
+                description: "True when the dry-run indicates the request is at or above the approval threshold.",
+            },
+        }, ["outcome", "explanation"]),
     },
     paybond_bootstrap_sandbox_guardrail: {
         title: "Bootstrap Sandbox Guardrail",
@@ -183,22 +242,89 @@ const TOOL_SELECTION_METADATA = {
     },
     paybond_get_reputation_receipt: {
         title: "Get Reputation Receipt",
+        description: "Use this when you need the signed Signal reputation receipt for one known tenant-scoped operator DID " +
+            "(score, metrics, reason codes, and Ed25519 signing material under receipt). " +
+            "Requires PAYBOND_API_KEY with Signal analytics read access. " +
+            "Do not use this for tenant-wide aggregates—call paybond_get_portfolio_summary—or a portable signed operator list—call " +
+            "paybond_get_signed_portfolio_artifact—or one operator's fraud review posture—call paybond_get_fraud_assessment. " +
+            "Idempotent read with no side effects; returns null when no receipt exists for that operator and score_version.",
         annotations: readOnlyToolAnnotations("Get Reputation Receipt"),
         outputSchema: outputObjectSchema({
-            tenant_id: { type: "string" },
-            operator_did: { type: "string" },
-            signature_hex: { type: "string" },
+            schema_version: {
+                type: "integer",
+                description: "Reputation receipt envelope schema version.",
+            },
+            updated_at: {
+                type: "string",
+                description: "RFC3339 timestamp when the stored receipt row was last updated.",
+            },
+            receipt: {
+                type: "object",
+                additionalProperties: true,
+                description: "Signed Signal receipt for the operator (tenant_id, operator_did, score_version, score, metrics, reason_codes, signing_algorithm, message_digest_hex, signing_public_key_hex, signature_hex).",
+                examples: [
+                    {
+                        tenant_id: "tenant-a",
+                        operator_did: "did:web:vendor.example#booker-agent",
+                        score_version: "1.0",
+                        score: 812,
+                        signature_hex: "ab".repeat(32),
+                    },
+                ],
+            },
         }),
     },
     paybond_get_portfolio_summary: {
         title: "Get Portfolio Summary",
+        description: "Use this when you need a read-only, tenant-scoped Signal portfolio aggregate for the authenticated API key " +
+            "(operator_count, average_score, total_terminal_intents, total_receipted_volume_cents, operators_under_review, " +
+            "and checkpoint_last_ledger_seq). Requires PAYBOND_API_KEY with Signal analytics read access and the private-dashboards feature. " +
+            "Do not use this when you need a portable signed operator list for partner or verifier sharing—call " +
+            "paybond_get_signed_portfolio_artifact instead—or for one operator's signed receipt—call paybond_get_reputation_receipt. " +
+            "Idempotent read with no side effects; auth, RBAC, feature, or gateway failures surface as tool errors.",
         annotations: readOnlyToolAnnotations("Get Portfolio Summary"),
         outputSchema: outputObjectSchema({
-            tenant_id: { type: "string" },
-            score_model_version: { type: "string" },
-            operators: {
-                type: "array",
-                items: { type: "object", additionalProperties: true },
+            schema_version: {
+                type: "integer",
+                description: "Portfolio summary schema version (currently 1).",
+            },
+            tenant_id: {
+                type: "string",
+                description: "Tenant echoed by the gateway for the authenticated API key (example: tenant-a).",
+                examples: ["tenant-a"],
+            },
+            score_model_version: {
+                type: "string",
+                description: "Score model version used for the aggregate (echoes the requested score_version or the gateway default 1.0).",
+                examples: ["1.0"],
+            },
+            scoring_model: {
+                type: "string",
+                description: "Scoring model identifier used by Signal for this summary.",
+            },
+            checkpoint_last_ledger_seq: {
+                type: "integer",
+                description: "Last ledger sequence included in the tenant Signal checkpoint.",
+            },
+            operator_count: {
+                type: "integer",
+                description: "Number of operators with reputation data for this score model version.",
+            },
+            average_score: {
+                type: "number",
+                description: "Average operator score across the tenant portfolio for this score model version.",
+            },
+            total_terminal_intents: {
+                type: "integer",
+                description: "Aggregate terminal Harbor intents across operators in the portfolio.",
+            },
+            total_receipted_volume_cents: {
+                type: "integer",
+                description: "Aggregate receipted settlement volume in cents across the portfolio.",
+            },
+            operators_under_review: {
+                type: "integer",
+                description: "Count of operators currently under Signal review for this score model version.",
             },
         }),
     },
@@ -214,21 +340,106 @@ const TOOL_SELECTION_METADATA = {
     },
     paybond_get_fraud_assessment: {
         title: "Get Fraud Assessment",
+        description: "Use this when you need the read-only fraud assessment and review posture for one known tenant-scoped operator DID (review state, fraud signals, and compact fraud_assessment). " +
+            "Example: look up operator_did=did:web:vendor.example#booker-agent (optionally score_version=1.0) before deciding whether to continue a spend workflow for that operator. " +
+            "Do not use this for tenant-wide fraud backtesting metrics—call paybond_get_fraud_metrics instead—or for Harbor intent escrow detail—call paybond_get_intent. " +
+            "Idempotent read; returns null when no assessment exists for that operator.",
         annotations: readOnlyToolAnnotations("Get Fraud Assessment"),
         outputSchema: outputObjectSchema({
-            tenant_id: { type: "string" },
-            operator_did: { type: "string" },
-            fraud_assessment: { type: "object", additionalProperties: true },
+            tenant_id: {
+                type: "string",
+                description: "Tenant echoed by the gateway for the authenticated API key (example: tenant-a).",
+                examples: ["tenant-a"],
+            },
+            operator_did: {
+                type: "string",
+                description: "Operator DID echoed from the assessment response (example: did:web:vendor.example#booker-agent).",
+                examples: ["did:web:vendor.example#booker-agent"],
+            },
+            fraud_assessment: {
+                type: "object",
+                additionalProperties: true,
+                description: "Compact fraud assessment for the operator (level, severity, signal counts, summary). Example shape: {\"level\":\"high\",\"highest_severity\":\"high\",\"signal_count\":1,\"summary\":\"level=high\"}.",
+                examples: [
+                    {
+                        level: "high",
+                        highest_severity: "high",
+                        signal_count: 1,
+                        summary: "level=high",
+                    },
+                ],
+            },
         }),
     },
     paybond_get_fraud_metrics: {
         title: "Get Fraud Metrics",
+        description: "Use this when you need tenant-wide Signal fraud backtesting and monitoring metrics over a rolling window " +
+            "(flagged operators, severity counts, review outcomes, precision/false-positive rates, and backtest_summary). " +
+            "Requires PAYBOND_API_KEY with Signal analytics read access and the private-dashboards feature. " +
+            "Do not use this for one operator's fraud posture—call paybond_get_fraud_assessment instead—or for Harbor intent escrow detail—call paybond_get_intent. " +
+            "Idempotent read with no side effects; omit window to default to 24h; unsupported windows fail with HTTP 400 " +
+            "(\"window must be one of 24h, 7d, or 30d\").",
         annotations: readOnlyToolAnnotations("Get Fraud Metrics"),
         outputSchema: outputObjectSchema({
-            tenant_id: { type: "string" },
-            window: { type: "string" },
-            flagged_operator_count: { type: "integer" },
-            critical_signal_count: { type: "integer" },
+            tenant_id: {
+                type: "string",
+                description: "Tenant echoed by the gateway for the authenticated API key (example: tenant-a).",
+                examples: ["tenant-a"],
+            },
+            score_model_version: {
+                type: "string",
+                description: "Score model version used for the metrics (echoes the requested score_version or the gateway default 1.0).",
+                examples: ["1.0"],
+            },
+            window: {
+                type: "string",
+                description: "Active metrics window label: 24h, 7d, or 30d.",
+                examples: ["24h", "7d", "30d"],
+            },
+            window_started_at: {
+                type: "string",
+                description: "RFC3339 start of the evaluated rolling window.",
+            },
+            window_ended_at: {
+                type: "string",
+                description: "RFC3339 end of the evaluated rolling window.",
+            },
+            flagged_operator_count: {
+                type: "integer",
+                description: "Operators with at least one fraud signal in the window.",
+            },
+            critical_signal_count: {
+                type: "integer",
+                description: "Count of critical-severity fraud signals in the window.",
+            },
+            high_signal_count: {
+                type: "integer",
+                description: "Count of high-severity fraud signals in the window.",
+            },
+            elevated_signal_count: {
+                type: "integer",
+                description: "Count of elevated-severity fraud signals in the window.",
+            },
+            review_open_count: {
+                type: "integer",
+                description: "Operators currently in an open review state.",
+            },
+            labeled_outcome_count: {
+                type: "integer",
+                description: "Review outcomes labeled in the window (confirmed risk, false positive, or needs more evidence).",
+            },
+            confirmed_risk_count: {
+                type: "integer",
+                description: "Labeled confirmed-risk outcomes in the window.",
+            },
+            false_positive_count: {
+                type: "integer",
+                description: "Labeled false-positive outcomes in the window.",
+            },
+            backtest_summary: {
+                type: "string",
+                description: "Human-readable backtest summary derived from the window metrics.",
+            },
         }),
     },
     paybond_get_a2a_agent_card: {
@@ -299,10 +510,40 @@ const TOOL_SELECTION_METADATA = {
     },
     paybond_verify_protocol_receipt_v1: {
         title: "Verify Protocol Receipt",
+        description: "Use this when you already have a signed protocol-v2 authorization or settlement receipt JSON object and need offline Ed25519 verification (structure, message digest, and signature) through the gateway. " +
+            "Do not use this to verify AgentMandateV1 envelopes—call paybond_verify_agent_mandate_v1—or to check a Harbor capability token before spend—call paybond_verify_capability or paybond_authorize_agent_spend. " +
+            "To load a settlement receipt by intent UUID first, call paybond_get_settlement_receipt_v1 then pass its body here. " +
+            "Read-only and side-effect free: success returns valid=true with kind, receipt_id, tenant_id, and the normalized receipt; unsupported kind, malformed JSON, digest mismatch, or bad signature fail with a gateway error (typically HTTP 400).",
         annotations: readOnlyToolAnnotations("Verify Protocol Receipt"),
         outputSchema: outputObjectSchema({
-            valid: { type: "boolean" },
-            receipt_id: { type: "string" },
+            valid: {
+                type: "boolean",
+                description: "True when the gateway accepted the receipt structure and Ed25519 signature. Example: true.",
+                examples: [true],
+            },
+            kind: {
+                type: "string",
+                description: "Verified receipt kind echoed from the normalized receipt. One of paybond.protocol_authorization_receipt_v1 or paybond.protocol_settlement_receipt_v1.",
+                examples: [
+                    "paybond.protocol_authorization_receipt_v1",
+                    "paybond.protocol_settlement_receipt_v1",
+                ],
+            },
+            receipt_id: {
+                type: "string",
+                description: "Canonical receipt identifier from the verified receipt.",
+                examples: ["550e8400-e29b-41d4-a716-446655440000"],
+            },
+            tenant_id: {
+                type: "string",
+                description: "Tenant id embedded in the verified receipt (not invented by the caller).",
+                examples: ["acme-pilot"],
+            },
+            receipt: {
+                type: "object",
+                additionalProperties: true,
+                description: "Normalized verified receipt object matching the input kind (authorization or settlement fields plus signing material).",
+            },
         }),
     },
     paybond_create_intent: {
@@ -732,6 +973,70 @@ class PaybondMCPRuntime {
         }
         return body;
     }
+    /**
+     * Side-effect-free spend policy dry-run via `POST /v1/spend/preflight`.
+     * Tenant scope comes from the authenticated API key only.
+     */
+    async spendPreflight(init) {
+        const payload = {
+            intent_id: init.intentId,
+            operation: init.operation?.trim() || "*",
+            requested_spend_cents: init.requestedSpendCents ?? 0,
+        };
+        if (init.vendorId?.trim()) {
+            payload.vendor_id = init.vendorId.trim();
+        }
+        if (init.toolName?.trim()) {
+            payload.tool_name = init.toolName.trim();
+        }
+        if (init.taskId?.trim()) {
+            payload.task_id = init.taskId.trim();
+        }
+        if (init.workflowId?.trim()) {
+            payload.workflow_id = init.workflowId.trim();
+        }
+        if (init.toolCallId?.trim()) {
+            payload.tool_call_id = init.toolCallId.trim();
+        }
+        if (init.currency?.trim()) {
+            payload.currency = init.currency.trim();
+        }
+        if (init.agentSubject?.trim()) {
+            payload.agent_subject = init.agentSubject.trim();
+        }
+        if (init.approvalToken?.trim()) {
+            payload.approval_token = init.approvalToken.trim();
+        }
+        return this.gateway.postJSON("/v1/spend/preflight", payload, {
+            "x-tenant-id": await this.tenantId(),
+        });
+    }
+    async getBudgetRemaining(init) {
+        const body = await this.spendPreflight(init);
+        return {
+            remaining_cents: body.remaining_cents ?? null,
+            spend_scope: body.spend_scope ?? null,
+            policy_version: body.policy_version ?? null,
+        };
+    }
+    async explainPolicy(init) {
+        const body = await this.spendPreflight(init);
+        const reasonCodes = Array.isArray(body.reason_codes)
+            ? body.reason_codes.map((code) => String(code))
+            : [];
+        const outcome = normalizeExplainPolicyOutcome(String(body.outcome ?? ""), String(body.classification ?? ""));
+        const result = {
+            outcome,
+            reason_codes: reasonCodes,
+            explanation: String(body.explanation ?? ""),
+            remaining_cents: body.remaining_cents ?? null,
+        };
+        if (reasonCodes.includes("approval_threshold_exceeded") ||
+            outcome === "approval_required") {
+            result.approval_threshold_exceeded = reasonCodes.includes("approval_threshold_exceeded");
+        }
+        return result;
+    }
     async bootstrapSandboxGuardrail(init) {
         const expectedTenant = await this.tenantId();
         const payload = {
@@ -904,11 +1209,17 @@ export function formatMcpStdioFrame(response) {
     const body = JSON.stringify(response);
     return `Content-Length: ${Buffer.byteLength(body, "utf8")}\r\n\r\n${body}`;
 }
+/** Newline-delimited JSON-RPC frame used by mcp-proxy and recent MCP SDKs. */
+export function formatMcpNdjsonFrame(response) {
+    return `${JSON.stringify(response)}\n`;
+}
 export class PaybondMCPServer {
     runtime;
     tools;
     toolPolicy;
     initialized = false;
+    /** Framing negotiated from the first successfully parsed stdin message. */
+    stdioFraming = "content-length";
     constructor(settings) {
         if (!settings.apiKey.trim()) {
             throw new Error("PAYBOND_API_KEY is required");
@@ -1109,7 +1420,11 @@ export class PaybondMCPServer {
     runStdio() {
         let buffer = Buffer.alloc(0);
         process.stdin.on("data", (chunk) => {
-            buffer = Buffer.concat([buffer, Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk, "utf8")]);
+            buffer = Buffer.concat([
+                buffer,
+                Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk, "utf8"),
+            ]);
+            // Prefer Content-Length frames when present (LSP-style MCP hosts).
             while (true) {
                 const headerEnd = buffer.indexOf("\r\n\r\n");
                 if (headerEnd < 0) {
@@ -1131,8 +1446,36 @@ export class PaybondMCPServer {
                 }
                 const body = buffer.subarray(bodyStart, frameEnd).toString("utf8");
                 buffer = buffer.subarray(frameEnd);
+                this.stdioFraming = "content-length";
                 void this.handleLine(body);
             }
+            // Do not let NDJSON parsing consume an in-progress Content-Length frame.
+            const preview = buffer
+                .subarray(0, Math.min(buffer.length, 64))
+                .toString("utf8")
+                .trimStart()
+                .toLowerCase();
+            if (preview.startsWith("content-length:")) {
+                return;
+            }
+            // Newline-delimited JSON-RPC (mcp-proxy / recent MCP SDKs).
+            while (true) {
+                const newline = buffer.indexOf("\n");
+                if (newline < 0) {
+                    break;
+                }
+                const rawLine = buffer
+                    .subarray(0, newline)
+                    .toString("utf8")
+                    .replace(/\r$/, "");
+                buffer = buffer.subarray(newline + 1);
+                const line = rawLine.trim();
+                if (!line.startsWith("{")) {
+                    continue;
+                }
+                this.stdioFraming = "ndjson";
+                void this.handleLine(line);
+            }
         });
         process.stdin.on("end", () => {
             if (buffer.length > 0) {
@@ -1156,7 +1499,10 @@ export class PaybondMCPServer {
         }
     }
     writeResponse(response) {
-        process.stdout.write(formatMcpStdioFrame(response));
+        const frame = this.stdioFraming === "ndjson"
+            ? formatMcpNdjsonFrame(response)
+            : formatMcpStdioFrame(response);
+        process.stdout.write(frame);
     }
     buildTools(settings) {
         const tools = [
@@ -1234,6 +1580,18 @@ export class PaybondMCPServer {
                     });
                 },
             },
+            {
+                name: "paybond_get_budget_remaining",
+                description: "Read-only dry-run of remaining spend budget for a tenant-bound intent via gateway spend preflight.",
+                inputSchema: spendPreflightInputSchema(),
+                call: async (args) => this.runtime.getBudgetRemaining(parseSpendPreflightArgs(args)),
+            },
+            {
+                name: "paybond_explain_policy",
+                description: "Read-only dry-run explanation of whether proposed spend would allow, require approval, or deny under tenant spend policy.",
+                inputSchema: spendPreflightInputSchema(),
+                call: async (args) => this.runtime.explainPolicy(parseSpendPreflightArgs(args)),
+            },
             {
                 name: "paybond_bootstrap_sandbox_guardrail",
                 description: "Bootstrap a sandbox-only Paybond guardrail intent for a first paid-tool integration. Tenant scope is derived from the configured service-account API key and the route never touches live settlement rails.",
@@ -1392,8 +1750,19 @@ export class PaybondMCPServer {
                 name: "paybond_get_reputation_receipt",
                 description: "Fetch the signed Signal receipt for one operator DID.",
                 inputSchema: objectSchema({
-                    operator_did: { type: "string" },
-                    score_version: { type: "string" },
+                    operator_did: {
+                        type: "string",
+                        description: "Tenant-scoped operator DID whose signed Signal reputation receipt to fetch. Must belong to the authenticated tenant; do not invent tenant identifiers. Examples: did:web:vendor.example#booker-agent, did:key:z6MkhaXgBZDvotDkL5257faiztiGiC2QtKLGpbnnEGta2doK.",
+                        examples: [
+                            "did:web:vendor.example#booker-agent",
+                            "did:key:z6MkhaXgBZDvotDkL5257faiztiGiC2QtKLGpbnnEGta2doK",
+                        ],
+                    },
+                    score_version: {
+                        type: "string",
+                        description: "Optional Signal score model version to query. Omit to use the gateway default current model (1.0). Example: 1.0.",
+                        examples: ["1.0"],
+                    },
                 }, ["operator_did"]),
                 call: async (args) => (await this.runtime.signal()).getReputationReceipt(stringArg(args.operator_did, "operator_did"), optionalString(args.score_version)),
             },
@@ -1401,7 +1770,11 @@ export class PaybondMCPServer {
                 name: "paybond_get_portfolio_summary",
                 description: "Fetch the tenant-scoped Signal portfolio summary.",
                 inputSchema: objectSchema({
-                    score_version: { type: "string" },
+                    score_version: {
+                        type: "string",
+                        description: "Optional Signal score model version to query. Omit to use the gateway default current model (1.0). Example: 1.0.",
+                        examples: ["1.0"],
+                    },
                 }),
                 call: async (args) => (await this.runtime.signal()).getPortfolioSummary(optionalString(args.score_version)),
             },
@@ -1417,8 +1790,19 @@ export class PaybondMCPServer {
                 name: "paybond_get_fraud_assessment",
                 description: "Fetch the read-only fraud assessment for one tenant-scoped operator DID.",
                 inputSchema: objectSchema({
-                    operator_did: { type: "string" },
-                    score_version: { type: "string" },
+                    operator_did: {
+                        type: "string",
+                        description: "Tenant-scoped operator DID to assess. Must belong to the authenticated tenant; do not invent tenant identifiers. Examples: did:web:vendor.example#booker-agent, did:key:z6MkhaXgBZDvotDkL5257faiztiGiC2QtKLGpbnnEGta2doK.",
+                        examples: [
+                            "did:web:vendor.example#booker-agent",
+                            "did:key:z6MkhaXgBZDvotDkL5257faiztiGiC2QtKLGpbnnEGta2doK",
+                        ],
+                    },
+                    score_version: {
+                        type: "string",
+                        description: "Optional Signal score model version to query. Omit to use the gateway default current model. Example: 1.0.",
+                        examples: ["1.0"],
+                    },
                 }, ["operator_did"]),
                 call: async (args) => (await this.runtime.fraud()).getFraudAssessment(stringArg(args.operator_did, "operator_did"), optionalString(args.score_version)),
             },
@@ -1426,8 +1810,17 @@ export class PaybondMCPServer {
                 name: "paybond_get_fraud_metrics",
                 description: "Fetch tenant-scoped read-only fraud backtesting and monitoring metrics for a supported active window.",
                 inputSchema: objectSchema({
-                    window: { type: "string", enum: ["24h", "7d", "30d"] },
-                    score_version: { type: "string" },
+                    window: {
+                        type: "string",
+                        enum: ["24h", "7d", "30d"],
+                        description: "Rolling metrics window. Allowed values: 24h, 7d, 30d. Omit to use the gateway default 24h. Unsupported values fail with HTTP 400.",
+                        examples: ["24h", "7d", "30d"],
+                    },
+                    score_version: {
+                        type: "string",
+                        description: "Optional Signal score model version to query. Omit to use the gateway default current model (1.0). Example: 1.0.",
+                        examples: ["1.0"],
+                    },
                 }),
                 call: async (args) => (await this.runtime.fraud()).getFraudMetrics({
                     window: optionalString(args.window),
@@ -1505,9 +1898,16 @@ export class PaybondMCPServer {
             },
             {
                 name: "paybond_verify_protocol_receipt_v1",
-                description: "Verify a protocol-v2 authorization or settlement receipt through the gateway.",
+                description: "Verify a signed protocol-v2 authorization or settlement receipt through the gateway.",
                 inputSchema: objectSchema({
-                    receipt: { type: "object", additionalProperties: true },
+                    receipt: {
+                        type: "object",
+                        additionalProperties: true,
+                        description: "Complete signed protocol receipt object posted as the verify request body (not a receipt_id string). " +
+                            "Discriminate on kind: paybond.protocol_authorization_receipt_v1 requires schema_version=1, receipt_version=\"1\", receipt_id, issued_at, status (authorized), intent_id, tenant_id, verifier_id, transport_binding, mandate_digest_sha256_hex, imported_mandate_signing_public_key_ed25519_hex, authorization, agent, allowed_actions, allowed_tools, spend_ceiling, settlement, constraint, expires_at, nonce, human_presence_mode, plus signing_algorithm=ed25519-sha256-json-v1, message_digest_sha256_hex, signing_public_key_ed25519_hex, and ed25519_signature_hex. " +
+                            "paybond.protocol_settlement_receipt_v1 requires schema_version=1, receipt_version=\"1\", receipt_id, issued_at, intent_id, tenant_id, verifier_id, transport_binding, authorization_receipt_id, mandate_digest_sha256_hex, harbor_state, settlement_rail, settlement_mode, principal_did, payee_did, currency, amount_cents, terminal_observed_at, optional predicate_passed, and the same Ed25519 signing fields. " +
+                            "Obtain receipts from mandate import, paybond_get_settlement_receipt_v1, audit export, or partner handoff—do not invent digests or signatures.",
+                    },
                 }, ["receipt"]),
                 call: async (args) => this.runtime.verifyProtocolReceiptV1(ensureObject(args.receipt, "receipt")),
             },
@@ -1681,6 +2081,93 @@ function stringArg(value, field) {
 function optionalString(value) {
     return typeof value === "string" && value.trim() ? value.trim() : undefined;
 }
+function spendPreflightInputSchema() {
+    return objectSchema({
+        intent_id: {
+            type: "string",
+            description: "Canonical Harbor intent UUID.",
+        },
+        operation: {
+            type: "string",
+            description: "Optional delegated operation or tool name. Defaults to * when omitted.",
+        },
+        requested_spend_cents: {
+            type: "integer",
+            description: "Optional proposed spend in cents for the dry-run evaluation.",
+        },
+        vendor_id: {
+            type: "string",
+            description: "Optional vendor scope hint for policy evaluation.",
+        },
+        tool_name: {
+            type: "string",
+            description: "Optional tool name scope hint for policy evaluation.",
+        },
+        task_id: {
+            type: "string",
+            description: "Optional task scope hint for policy evaluation.",
+        },
+        workflow_id: {
+            type: "string",
+            description: "Optional workflow scope hint for policy evaluation.",
+        },
+        tool_call_id: {
+            type: "string",
+            description: "Optional tool-call correlation id for policy evaluation.",
+        },
+        currency: {
+            type: "string",
+            description: "Optional ISO currency code for the proposed spend.",
+        },
+        agent_subject: {
+            type: "string",
+            description: "Optional agent subject for agent-scoped caps.",
+        },
+        approval_token: {
+            type: "string",
+            description: "Optional approval token to evaluate against pending approval state (not consumed).",
+        },
+    }, ["intent_id"]);
+}
+function parseSpendPreflightArgs(args) {
+    return {
+        intentId: uuidArg(args.intent_id, "intent_id"),
+        operation: optionalString(args.operation),
+        requestedSpendCents: args.requested_spend_cents === undefined
+            ? undefined
+            : intArg(args.requested_spend_cents, "requested_spend_cents"),
+        vendorId: optionalString(args.vendor_id),
+        toolName: optionalString(args.tool_name),
+        taskId: optionalString(args.task_id),
+        workflowId: optionalString(args.workflow_id),
+        toolCallId: optionalString(args.tool_call_id),
+        currency: optionalString(args.currency),
+        agentSubject: optionalString(args.agent_subject),
+        approvalToken: optionalString(args.approval_token),
+    };
+}
+/** Normalize gateway preflight outcomes to the agent-facing explain-policy set. */
+function normalizeExplainPolicyOutcome(outcome, classification) {
+    const normalized = outcome.trim().toLowerCase();
+    if (normalized === "allow" || normalized === "anomaly_observe") {
+        return "allow";
+    }
+    if (normalized === "approval_required" ||
+        normalized === "anomaly_escalate") {
+        return "approval_required";
+    }
+    if (normalized === "deny") {
+        return "deny";
+    }
+    const classNorm = classification.trim().toLowerCase();
+    if (classNorm === "allow") {
+        return "allow";
+    }
+    if (classNorm === "hold") {
+        return "approval_required";
+    }
+    return "deny";
+}
 function intArg(value, field) {
     const parsed = typeof value === "number"
         ? value