@tokenzero/mcp 0.2.10 → 0.2.12

package/bin/tokenzero-mcp.mjs

@@ -9,7 +9,7 @@ import readline from "node:readline";
 import crypto from "node:crypto";
 
 // src/pricing.ts
-var PRICING_VERSION = "2026-10-28";
+var PRICING_VERSION = "2026-11-01";
 var P = (provider, tier, input, output, cachedInput, cacheWrite, extra = {}) => ({ provider, tier, input, output, cachedInput, cacheWrite, ...extra });
 var PRICES = {
   // ---- OpenAI (standard tier, ≤272K; cache write = 30m TTL rate) ----
@@ -23,6 +23,9 @@ var PRICES = {
   "gpt-5.6-luna": P("openai", "fast", 0.2, 1.2, 0.02, 0.25, { longCtxAt: 272e3, longCtx: { input: 0.4, output: 1.8, cachedInput: 0.04, cacheWrite: 0.5 }, minCacheTokens: 1024 }),
   // Daybreak cyber model: access needs OpenAI approval; 400K window
   "gpt-5.6-cyber": P("openai", "frontier", 12.5, 75, 1.25, 15.625, { longCtxAt: 272e3, longCtx: { input: 25, output: 112.5, cachedInput: 2.5, cacheWrite: 31.25 }, minCacheTokens: 1024 }),
+  // 2026-11-01 (developers.openai.com/api/docs/pricing): Astra for Law at Astra rates (promo), Daybreak gpt-5.5-cyber.
+  "gpt-6-astra-law": P("openai", "frontier", 10, 50, 1, void 0, { minCacheTokens: 1024 }),
+  "gpt-5.5-cyber": P("openai", "frontier", 12.5, 75),
   "gpt-5.3-codex": P("openai", "mid", 1.75, 14, 0.175, void 0, { minCacheTokens: 1024 }),
   "gpt-5.5": P("openai", "frontier", 5, 30, 0.5, void 0, { fastFactor: 2.5, longCtxAt: 272e3, longCtx: { input: 10, output: 45, cachedInput: 1 }, minCacheTokens: 1024 }),
   "gpt-5.5-pro": P("openai", "frontier", 30, 180),
@@ -42,6 +45,13 @@ var PRICES = {
   "o3-mini": P("openai", "fast", 1.1, 4.4, 0.55, void 0, { minCacheTokens: 1024 }),
   "o4-mini": P("openai", "fast", 1.1, 4.4, 0.275, void 0, { minCacheTokens: 1024 }),
   "o1": P("openai", "frontier", 15, 60, 7.5),
+  "o1-pro": P("openai", "frontier", 150, 600),
+  // Legacy models still on the pricing page: audited repos use them, so receipts and findings price them correctly
+  // instead of falling back to the nearest prefix row (gpt-4o-2024-05-13 is $5/$15, not gpt-4o's $2.50/$10).
+  "gpt-4o-2024-05-13": P("openai", "mid", 5, 15),
+  "gpt-4-turbo": P("openai", "frontier", 10, 30),
+  "gpt-4-0613": P("openai", "frontier", 30, 60),
+  "gpt-3.5-turbo": P("openai", "fast", 0.5, 1.5),
   "gpt-4.1": P("openai", "mid", 2, 8, 0.5, void 0, { fastFactor: 1.75, minCacheTokens: 1024 }),
   "gpt-4.1-mini": P("openai", "fast", 0.4, 1.6, 0.1, void 0, { minCacheTokens: 1024 }),
   "gpt-4.1-nano": P("openai", "fast", 0.1, 0.4, 0.025, void 0, { minCacheTokens: 1024 }),
@@ -85,7 +95,7 @@ var PRICES = {
   "deepseek-v4-pro": P("deepseek", "mid", 1.32, 3.96, 0.044),
   "deepseek-flash": P("deepseek", "fast", 0.3, 1.2, 6e-3)
 };
-var MODEL_ALIASES = { "deepseek-v4-flash": "deepseek-flash", "deepseek-v4-flash-vision-exp": "deepseek-flash" };
+var MODEL_ALIASES = { "deepseek-v4-flash": "deepseek-flash", "deepseek-v4-flash-vision-exp": "deepseek-flash", "deepseek-v4.1-flash": "deepseek-flash", "deepseek-v4.1-flash-1017": "deepseek-flash", "deepseek-v4-pro-0813": "deepseek-v4-pro", "gpt-4-turbo-2024-04-09": "gpt-4-turbo", "gpt-4": "gpt-4-0613", "gpt-3.5-turbo-0125": "gpt-3.5-turbo" };
 var CASCADE_TARGET = {
   "gpt-6-astra": "gpt-6.1-sol",
   "gpt-6.1-sol": "gpt-6.1-terra",
@@ -123,6 +133,7 @@ var CASCADE_TARGET = {
   "gemini-2.5-flash": "gemini-2.5-flash-lite"
 };
 var KEYS_BY_LENGTH = Object.keys(PRICES).sort((a, b) => b.length - a.length);
+var DISTINCT_VARIANT = /^(instruct|audio|realtime|search|transcribe|tts|image|embedding|vision|computer-use|deep-research)\b/;
 function normalizeModel(model) {
   if (typeof model !== "string" || !model) return null;
   let m = model.toLowerCase().trim();
@@ -130,11 +141,12 @@ function normalizeModel(model) {
   m = m.replace(/^(us|eu|apac|global)\./, "").replace(/^(anthropic|amazon|meta|mistral|openai)\./, "");
   m = m.replace(/-v\d+(?::\d+)?$/, "");
   if (MODEL_ALIASES[m]) m = MODEL_ALIASES[m];
+  if (PRICES[m]) return m;
   m = m.replace(/@\d{8}$/, "");
   m = m.replace(/-(\d{8}|\d{4}-\d{2}-\d{2})$/, "");
   m = m.replace(/-(latest|preview|customtools|thinking)$/, "");
   if (PRICES[m]) return m;
-  for (const k of KEYS_BY_LENGTH) if (m.startsWith(k + "-") || m === k) return k;
+  for (const k of KEYS_BY_LENGTH) if (m === k || m.startsWith(k + "-") && !DISTINCT_VARIANT.test(m.slice(k.length + 1))) return k;
   return null;
 }
 var isFreeModel = (model) => typeof model === "string" && (/:free$/i.test(model.trim()) || /^openrouter\/free$/i.test(model.trim()));
@@ -920,12 +932,58 @@ function manifests(root) {
   });
 }
 
+// src/events-types.ts
+var EVENTS_EXTENSION = "io.modelcontextprotocol/events";
+var EVENT_ERR = { invalidParams: -32602, notFound: -32023, forbidden: -32024, resourceExhausted: -32025, unsupported: -32026 };
+var keyNote = "Needs a workspace key (x-tokenzero-key or TOKENZERO_KEY).";
+var EVENT_TYPES = [
+  {
+    name: "spend.threshold_crossed",
+    description: `Fires once per period when the workspace's metered LLM spend through TokenZero crosses a dollar threshold: wake an agent on a cost spike. ${keyNote}`,
+    delivery: ["poll"],
+    inputSchema: { type: "object", properties: { usd: { type: "number", exclusiveMinimum: 0, description: "threshold in USD of actual (post-savings) spend" }, window: { type: "string", enum: ["day", "month"], default: "day", description: "UTC calendar period the spend accumulates over" } }, required: ["usd"] },
+    payloadSchema: { type: "object", properties: { workspace: { type: "string" }, window: { type: "string" }, period: { type: "string" }, threshold_usd: { type: "number" }, spend_usd: { type: "number" }, crossed_at: { type: "string", format: "date-time" }, request_id: { type: "string" }, dashboard_url: { type: "string" } } }
+  },
+  {
+    name: "savings.daily_report",
+    description: `One event per completed UTC day with traffic: requests, list-price baseline, actual spend and signed savings. ${keyNote}`,
+    delivery: ["poll"],
+    inputSchema: { type: "object", properties: {} },
+    payloadSchema: { type: "object", properties: { workspace: { type: "string" }, date: { type: "string", format: "date" }, requests: { type: "integer" }, cache_hits: { type: "integer" }, baseline_usd: { type: "number" }, actual_usd: { type: "number" }, savings_usd: { type: "number" }, savings_pct: { type: "number" } } }
+  },
+  {
+    name: "upstream.auth_failed",
+    description: `Fires when a provider rejects the BYOK key a call carried (401/403), with the fix and a keyless sandbox link. ${keyNote}`,
+    delivery: ["poll"],
+    inputSchema: { type: "object", properties: { provider: { type: "string", description: "only this provider (openai, anthropic, gemini, ...)" } } },
+    payloadSchema: { type: "object", properties: { workspace: { type: "string" }, request_id: { type: "string" }, provider: { type: "string" }, status: { type: "integer" }, model: { type: ["string", "null"] }, at: { type: "string", format: "date-time" }, fix: { type: "string" }, keyless_sandbox: { type: "string" } } }
+  },
+  {
+    name: "pricing.changed",
+    description: "Fires when a list price in TokenZero's curated, provider-verified pricing table changes (one event per model and field). No key needed.",
+    delivery: ["poll"],
+    inputSchema: { type: "object", properties: { models: { type: "array", items: { type: "string" }, description: "only these model ids" } } },
+    payloadSchema: { type: "object", properties: { model: { type: "string" }, field: { type: "string", enum: ["input", "output", "cachedInput", "cacheWrite", "added", "removed"] }, old: { type: ["number", "null"] }, new: { type: ["number", "null"] }, unit: { type: "string" }, pricing_version: { type: "string" } } }
+  }
+];
+
 // mcp/src/main.ts
-var VERSION = "0.2.10";
+var VERSION = "0.2.12";
 var BAKED_URL = true ? "https://tokenzero.sh" : "__TOKENZERO_DEFAULT_URL__";
 var SUPPORTED = [...MCP_LEGACY_VERSIONS];
 var SERVER = { name: "tokenzero", version: VERSION };
-var CAPABILITIES = { tools: { listChanged: false }, prompts: { listChanged: false } };
+var CAPABILITIES = { tools: { listChanged: false }, prompts: { listChanged: false }, extensions: { [EVENTS_EXTENSION]: {} } };
+var LEGACY_CAPABILITIES = { tools: { listChanged: false }, prompts: { listChanged: false }, experimental: { [EVENTS_EXTENSION]: {} } };
+var listEvents = () => ({ events: EVENT_TYPES.map((e) => ({ ...e, delivery: [...e.delivery] })) });
+async function pollEvents(params) {
+  const base = serviceUrl();
+  if (!base) return { error: { code: EVENT_ERR.unsupported, message: "Unsupported", data: { reason: "set TOKENZERO_URL to poll events" } } };
+  const key = process.env.TOKENZERO_KEY ?? "";
+  const r = await fetch(base + "/v1/events/poll", { method: "POST", redirect: "manual", signal: AbortSignal.timeout(2e4), headers: { "content-type": "application/json", "user-agent": `tokenzero-mcp/${VERSION}`, ...key ? { "x-tokenzero-key": key } : {} }, body: JSON.stringify({ name: params?.name, arguments: params?.arguments, cursor: params?.cursor ?? null, maxAgeMs: params?.maxAgeMs, maxEvents: params?.maxEvents }) });
+  const j = await r.json().catch(() => null);
+  if (r.ok && j && Array.isArray(j.events)) return { result: j };
+  return { error: j?.error && typeof j.error.code === "number" ? j.error : { code: RPC.internal, message: `events/poll failed: HTTP ${r.status}` } };
+}
 var PROMPTS = [
   { name: "cut-my-llm-bill", title: "Cut my LLM bill", description: "Find token waste in this project (scanned locally, nothing uploaded) and show the top fixes with estimated savings. No API key needed.", arguments: [{ name: "repo", description: "owner/name of a public GitHub repo instead of the local project", required: false }] },
   { name: "price-my-workload", title: "Price my LLM workload", description: "Price a workload across models at list prices, with the cheaper sibling and the effect of prompt caching. No API key needed.", arguments: [{ name: "model", description: "model id, e.g. gpt-6.1-sol or claude-sonnet-5-5", required: true }, { name: "requests_per_month", description: "monthly request volume (default 10000)", required: false }] },
@@ -965,7 +1023,7 @@ var TOOLS = [
     name: "calculate_model_arbitrage",
     title: "Price a workload across models",
     annotations: { title: "Price a workload across models", readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false },
-    description: `No API key needed; only \`model\` is required (defaults 2,000 tokens in and 400 out). Price a workload at list prices (table ${PRICING_VERSION}) for the requested model, its cheaper sibling and any extra models, with and without prompt caching. Runs locally.`,
+    description: `No API key needed; only \`model\` is required (defaults 2,000 tokens in and 400 out). Price a workload at list prices for the requested model, its cheaper sibling and any extra models, with and without prompt caching. Runs locally.`,
     inputSchema: { type: "object", properties: { model: { type: "string", description: "model id, e.g. gpt-6.1-sol or claude-sonnet-5-5" }, input_tokens: { type: "number", description: "per request; default 2000" }, output_tokens: { type: "number", description: "per request; default 400" }, requests_per_month: { type: "number" }, static_tokens: { type: "number", description: "portion of input_tokens that is a fixed prefix (system + tools)" }, compare_to: { type: "array", items: { type: "string" } } }, required: ["model"], additionalProperties: false }
   },
   {
@@ -1092,6 +1150,16 @@ async function handle(m) {
           return reply(modernResult(promptMessages(String(m.params?.name ?? ""), m.params?.arguments), SERVER));
         case "tools/call":
           return reply(modernResult(await call(), SERVER));
+        case "events/list":
+          return reply(modernResult(listEvents(), SERVER, true));
+        case "events/poll": {
+          const r = await pollEvents(m.params);
+          return r.result ? reply({ ...r.result, resultType: "complete", _meta: modernResult({}, SERVER)._meta }) : failData(r.error.code, r.error.message, r.error.data);
+        }
+        case "events/stream":
+        case "events/subscribe":
+        case "events/unsubscribe":
+          return failData(EVENT_ERR.unsupported, "Unsupported", { feature: "deliveryMode", value: m.method === "events/stream" ? "push" : "webhook", supported: ["poll"] });
         default:
           if (typeof m.method === "string" && m.method.startsWith("notifications/")) return;
           return fail(RPC.methodNotFound, `method not found: ${m.method}`);
@@ -1100,7 +1168,7 @@ async function handle(m) {
     switch (m?.method) {
       case "initialize": {
         const want = m.params?.protocolVersion;
-        return reply({ protocolVersion: SUPPORTED.includes(want) ? want : SUPPORTED[0], capabilities: CAPABILITIES, serverInfo: SERVER, instructions: INSTRUCTIONS });
+        return reply({ protocolVersion: SUPPORTED.includes(want) ? want : SUPPORTED[0], capabilities: LEGACY_CAPABILITIES, serverInfo: SERVER, instructions: INSTRUCTIONS });
       }
       case "ping":
         return reply({});
@@ -1115,6 +1183,21 @@ async function handle(m) {
         return reply(promptMessages(String(m.params?.name ?? ""), m.params?.arguments));
       case "tools/call":
         return reply(await call());
+      case "events/list":
+        return reply(listEvents());
+      case "events/poll": {
+        const r = await pollEvents(m.params);
+        if (r.result) {
+          const { resultType, ...rest } = r.result;
+          void resultType;
+          return reply(rest);
+        }
+        return failData(r.error.code, r.error.message, r.error.data);
+      }
+      case "events/stream":
+      case "events/subscribe":
+      case "events/unsubscribe":
+        return failData(EVENT_ERR.unsupported, "Unsupported", { feature: "deliveryMode", value: m.method === "events/stream" ? "push" : "webhook", supported: ["poll"] });
       default:
         if (typeof m?.method === "string" && m.method.startsWith("notifications/")) return;
         return fail(RPC.methodNotFound, `method not found: ${m?.method}`);

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@tokenzero/mcp",
   "mcpName": "sh.tokenzero/mcp",
-  "version": "0.2.10",
+  "version": "0.2.12",
   "description": "TokenZero MCP server (stdio): audit LLM token waste in your project locally, price model alternatives, and make prompts cache-friendly from Claude Code, Cursor, Windsurf or Cline.",
   "type": "module",
   "license": "Apache-2.0",