@index9/mcp 6.2.0 → 6.4.0

package/dist/cli.js

@@ -219,8 +219,14 @@ Parameters:
 - expectedPromptTokens: Estimated prompt-token count for dryRun cost estimation; overrides the prompt-string heuristic. Use to model "what would N-token requests cost?" without pasting N tokens.
 - expectedCompletionTokens: Optional completion token estimate used by dryRun
 - maxTokens, systemPrompt, temperature, topP, seed, responseFormat, enforceJson, retries: Live-testing controls (ignored when dryRun=true)
+- stream: Use OpenRouter's SSE streaming so capacity/refusal errors surface in ~1s instead of waiting the full per-model timeout for an empty 200. Defaults to false.
+- firstTokenTimeoutMs: Streaming-only deadline for the first delta. Defaults to 10s. If the upstream sends no token within this window, the request aborts and returns failureReason="timeout". Ignored when stream=false.
+- providerSort: "throughput" | "price" | "latency" \u2014 opt-in OpenRouter provider routing. Defaults to OpenRouter's load-balanced choice.
+- providerOrder: ordered list of provider slugs (up to 8). Try these providers first before falling back. Useful for steering around an overloaded provider for a single model.
+- fallbackModels: ordered list of model ids (up to 5). OpenRouter automatically retries the request against the next id when the primary is unavailable. Use sparingly \u2014 a benchmark should usually test the model you asked for, not a substitute.
+- debug: When true, each result includes a \`debug\` field with the raw upstream finish_reason, error message, \`providerName\` (OpenRouter routing/fulfillment provider \u2014 can differ from the publisher, e.g. an anthropic/* model served via Google Vertex), \`modelPublisher\` (derived from the canonical id prefix \u2014 e.g. "anthropic", "x-ai"), refusal, and usage. Use to diagnose "missing assistant text" without re-running.
 
-Results (live): each result carries modelId (the id you passed), resolvedModelId (canonical id, present when the input was an alias), ok, response, latencyMs, tokens { prompt, completion }, cost (USD; live from OpenRouter when available, else estimated from cached pricing), and truncated=true when finish_reason is "length". On failure, results include \`error\` (free-form) plus \`failureReason\` ("insufficient_credits" | "model_unavailable" | "rate_limited" | "timeout" | "invalid_request" | "unknown") so callers can pick a retry strategy without parsing the error string.
+Results (live): each result carries modelId (the id you passed), resolvedModelId (canonical id, present when the input was an alias), ok, response, latencyMs, tokens { prompt, completion }, cost (USD; live from OpenRouter when available, else estimated from cached pricing), and truncated=true when finish_reason is "length". On failure, results include \`error\` (free-form) plus \`failureReason\` ("insufficient_credits" | "model_unavailable" | "rate_limited" | "capacity" | "timeout" | "invalid_request" | "unknown") so callers can pick a retry strategy without parsing the error string. \`capacity\` indicates the provider is overloaded \u2014 apply a longer backoff or set \`fallbackModels\` and retry. When \`debug: true\` is set, each result also carries a \`debug\` block with the upstream provider's diagnostic fields.
 
 Results (dryRun): each entry carries \`tokenCostUsd\`, \`requestCostUsd\`, \`totalCostUsd\` (matches \`estimatedCost\`, includes per-request fees), and \`estimatedCostBasis\` (same enum as compare_models.workloadCosts). Use find_models or get_models first to identify model ids.
 
@@ -239,13 +245,16 @@ var PARAM_DESCRIPTIONS = {
   excludeFree: `When true, exclude models with id ending in ':free'. Useful for sortBy=price (which would otherwise be dominated by free-tier preview models) and when you want a paid SLA. Default false.`,
   requireKeywordMatch: `When true, suppress weak vector-only results from semantic queries. If no candidate has a BM25 keyword hit, returns an empty page with meta.confidence='low' and meta.lowConfidenceReason \u2014 instead of returning misleading nearest-neighbor matches. Filter-only queries (sortBy=created or sortBy=price without q) ignore this flag. Default false.`,
   expectedPromptTokens: `Expected number of prompt tokens for dryRun cost estimation. When set, overrides the heuristic that counts characters from the literal \`prompt\` string \u2014 use this for capacity planning ("what would 6000-token reviews cost?") without pasting filler. If both are omitted, the prompt string is tokenized at ~4 chars/token.`,
-  expectedCompletionTokens: `Expected number of completion tokens for cost estimation (default: 256). Typical ranges: 100-500 for quick tests, 1000-2000 for code generation, 4000+ for long-form content. This is a heuristic \u2014 actual billed tokens may differ.`
+  expectedCompletionTokens: `Expected number of completion tokens for cost estimation (default: 256). Typical ranges: 100-500 for quick tests, 1000-2000 for code generation, 4000+ for long-form content. This is a heuristic \u2014 actual billed tokens may differ.`,
+  minPrice: `Minimum effective prompt price in USD per million tokens. Matches the basis exposed in \`pricing.effectivePromptPerMillion\` and \`pageInfo.priceSortBasis\` \u2014 token price plus per-request fees scaled to per-million, so a model with $0.30/M tokens and a $0.30 per-request fee evaluates as $0.60/M effective. Models with no resolvable price (zero-token, no per-request fee, non-\`:free\` id) are excluded when this bound is set.`,
+  maxPrice: `Maximum effective prompt price in USD per million tokens. Matches the basis exposed in \`pricing.effectivePromptPerMillion\` and \`pageInfo.priceSortBasis\` \u2014 token price plus per-request fees scaled to per-million, so a model with $0.30/M tokens and a $0.30 per-request fee evaluates as $0.60/M effective and is excluded by \`maxPrice: 0.55\`. Models with no resolvable price (zero-token, no per-request fee, non-\`:free\` id) are excluded when this bound is set.`
 };
 var SITE = {
   nav: {
     brand: "index9",
     tools: "Tools",
     howItWorks: "How it works",
+    caseStudy: "Case study",
     install: "Install",
     faq: "FAQ",
     github: "GitHub",
@@ -253,26 +262,29 @@ var SITE = {
     installCta: "Install"
   },
   hero: {
-    titleLine1: "Pick the right AI model",
-    titleLine2: "from chat",
-    subtitle: "Index9 is an MCP server. Your coding assistant uses it to search, compare, and live-test 300+ models on the task you're working on, so it recommends the best fit.",
-    proof: ["Live OpenRouter data \xB7 300+ models \xB7 refreshed every 30 min"],
+    title: "Pick AI models from live data, not your assistant's memory.",
+    subtitle: "An MCP server with five tools for Claude Code, Cursor, VS Code, and Codex. Search, compare, cost-model, and live-test 300+ models.",
     pricingNote: "Free. You only pay OpenRouter for live model calls.",
-    getStarted: "Add index9 to your editor",
     seeHowItWorks: "See a real session",
-    updatedBadge: "OpenRouter data \xB7 refreshed "
+    updatedBadge: "OpenRouter data \xB7 refreshed ",
+    toolList: [
+      { name: "list_facets", role: "vocabulary" },
+      { name: "find_models", role: "search & filter" },
+      { name: "get_models", role: "full specs" },
+      { name: "compare_models", role: "side-by-side diff" },
+      { name: "test_model", role: "live inference" }
+    ]
   },
   problem: {
     label: "Why this exists",
     heading: "Your assistant's model knowledge is stale",
     body: [
       'New models ship every week. Pricing changes. "Use GPT-4" or "use Claude 3.5" is usually months behind reality.',
-      "Without live data, your assistant defaults to whatever it learned in training \u2014 often a model that's been superseded by something cheaper or better-suited to your task.",
-      "Index9 gives it the data and the tools to actually compare."
+      "Without live data, your assistant defaults to whatever it learned in training. Usually a model superseded by something cheaper or better-suited to your task.",
+      "Index9 gives it the data, and the tools to compare."
     ]
   },
   howItWorks: {
-    label: "How it works",
     heading: "How it works",
     subtitle: "Index9 adds 5 tools to your editor. Your assistant calls them when you ask about models.",
     steps: [
@@ -284,90 +296,95 @@ var SITE = {
       {
         number: "2",
         title: "Your assistant calls index9",
-        body: "It searches live model data, compares finalists, and runs your prompt against the top candidates."
+        body: "It searches live model data, compares finalists, and runs your prompt against the top picks."
       },
       {
         number: "3",
         title: "You get a measured pick",
-        body: "Backed by real cost numbers and real outputs \u2014 not training-data memory."
+        body: "Backed by real cost numbers and real outputs, not training-data memory."
       }
     ]
   },
   caseStudy: {
-    label: "Case study",
+    label: "Real session",
     heading: "A real session, not a mockup",
-    subheading: "A Claude Code session picking a TypeScript code-review model. Real tool calls, real verdict.",
+    subheading: "Claude Sonnet 4.6 driven through the index9 MCP server, picking a model for a TypeScript code-review bot. Real prompt, real tool calls, real verdict.",
     prompt: {
       title: "The prompt",
-      body: "Pick a model for a TypeScript code-review bot that runs on every PR. I want real quality without paying frontier rates on routine reviews. Test against this sample diff."
+      body: "Pick a model for a TypeScript code-review bot that runs on every PR. I want real quality without paying frontier rates. Quote model ids verbatim from the tool responses."
     },
     toolCalls: {
       title: "What the assistant did",
-      subtitle: "in order",
+      subtitle: "4 calls, 42s",
       calls: [
-        { tool: "find_models", params: "newest first", note: "skip stale training picks" },
         {
           tool: "find_models",
-          params: "code review + structured output",
+          params: "sortBy=created, function_calling + structured_output",
+          note: "skip stale training picks"
+        },
+        {
+          tool: "find_models",
+          params: "code review, maxPrice=$5/M, minContext=64K",
           note: "task fit"
         },
-        { tool: "find_models", params: "max $2/M, every-PR budget", note: "rule out frontier" },
-        { tool: "get_models", params: "8 candidates", note: "metadata lookup" },
         {
           tool: "compare_models",
-          params: "4 finalists, ~3000 token PR diff",
-          note: "per-PR cost projection"
+          params: "4 finalists, 3,000 prompt + 800 completion",
+          note: "per-PR cost diff"
         },
-        { tool: "test_model", params: "dry-run \xD7 4", note: "cost estimate" },
         {
           tool: "test_model",
-          params: "live \xD7 4, JSON output",
-          note: "real bug-catch test"
+          params: "dry-run \xD7 4, same token budget",
+          note: "cost confirmation"
         }
       ]
     },
-    consideredTitle: "Recent models, evaluated",
-    consideredSubtitle: "A trimmed view of the candidates the assistant ruled in and out. Each row pairs a decision with the reason behind it.",
+    consideredTitle: "Candidates the session evaluated",
+    consideredSubtitle: "Lifted from the tool responses. Per-PR cost is the dry-run test_model output for a 3,000 token diff and an 800 token review.",
     consideredRows: [
       {
-        id: "openai/gpt-5.5",
-        age: "1d ago",
+        id: "anthropic/claude-opus-4.7-fast",
+        age: "5d ago",
         decision: "skip",
-        reason: "~$0.027 per PR, 5\xD7 the pick for the same outcome"
+        reason: "$30/M input, filtered out by the maxPrice constraint on the second find_models call"
       },
       {
-        id: "xiaomi/mimo-v2.5-pro",
-        age: "3d ago",
+        id: "mistralai/devstral-medium",
+        age: "10mo ago",
         decision: "tested",
-        reason: "good structure, missed the precision edge case"
+        reason: "$0.40/M input, completion-heavy at $2/M, $0.0028 per PR"
       },
       {
-        id: "deepseek/deepseek-v4-flash",
-        age: "1d ago",
+        id: "qwen/qwen3-coder",
+        age: "10mo ago",
         decision: "tested",
-        reason: "7\xD7 cheaper than the pick, but missed both bugs"
+        reason: "$0.22/M input, 1M context but no file_input, $0.0021 per PR"
       },
       {
-        id: "z-ai/glm-5.1",
-        age: "2w ago",
+        id: "google/gemini-3.1-flash-lite",
+        age: "1w ago",
+        decision: "tested",
+        reason: "$0.25/M input, newest of the shortlist, adds reasoning and vision, $0.00195 per PR"
+      },
+      {
+        id: "mistralai/codestral-2508",
+        age: "9mo ago",
         decision: "shortlisted",
-        reason: "caught both bugs at ~$0.005 per PR"
+        reason: "$0.30/M input, Mistral's coding-specific model, $0.00162 per PR"
       }
     ],
     verdict: {
       title: "The pick",
-      model: "z-ai/glm-5.1",
-      body: "Open-weight, $1.05 per million input tokens. Caught both bugs in the sample diff at roughly $0.005 per PR, about 5\xD7 cheaper than running gpt-5.5 on every commit."
-    },
-    quote: {
-      body: "The frontier model would have caught both bugs, at 5\xD7 the cost. The cheapest candidate missed them entirely. Only the live test surfaced the model that did both.",
-      attribution: "index9 session trace"
+      model: "mistralai/codestral-2508",
+      body: "Lowest realistic per-PR cost at $0.00162 on a 3,000 token diff and an 800 token review. Mistral's coding-specific model, structured output, file input for diffs, 256K context. About 100\xD7 cheaper on input than running claude-opus-4.7-fast on every commit."
     }
   },
   toolsSection: {
     label: "Tools",
     heading: "The 5 tools",
     subheading: "Your assistant chains these together. You don't call them directly.",
+    keyNotePrefix: "Only",
+    keyNoteSuffix: "needs an OpenRouter key. The rest work out of the box.",
     openRouterKey: "OpenRouter API key",
     noKeyRequired: "No key required",
     requiresLabel: "Requires ",
@@ -404,7 +421,7 @@ var SITE = {
         action: "compare_models",
         displayName: "compare_models",
         fullName: null,
-        description: "Diffs 2\u201310 finalists side-by-side. Flags the cheapest pick for your expected token mix.",
+        description: "Diffs 2\u201310 finalists side-by-side. Flags the cheapest for your token mix.",
         badge: null,
         requiresKey: false
       },
@@ -413,7 +430,7 @@ var SITE = {
         action: "test_model",
         displayName: "test_model",
         fullName: null,
-        description: "Runs your prompt across models. Returns output, latency, and real cost. Or dry-run for cost only.",
+        description: "Runs your prompt across models. Returns output, latency, cost. Dry-run for cost only.",
         badge: "Live",
         requiresKey: true
       }
@@ -438,7 +455,7 @@ var SITE = {
       },
       {
         question: "Does it pick the model for me?",
-        answer: "No \u2014 it gives your assistant the data (search results, specs, cost diffs, live test outputs). Your assistant makes the call.",
+        answer: "No. It gives your assistant the data: search results, specs, cost diffs, live test outputs. Your assistant makes the call.",
         link: null
       },
       {
@@ -448,7 +465,7 @@ var SITE = {
       },
       {
         question: "Which models?",
-        answer: `${MODEL_COUNT} from OpenRouter \u2014 OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, and more. Metadata refreshes every 30 minutes.`,
+        answer: `${MODEL_COUNT} from OpenRouter: OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, and more. Metadata refreshes every 30 minutes.`,
         link: null
       },
       {
@@ -458,7 +475,7 @@ var SITE = {
       },
       {
         question: "What's the project status?",
-        answer: "Stable and in active use. Issues and feature requests welcome on GitHub.",
+        answer: "Stable. Issues and feature requests on GitHub.",
         link: null
       }
     ]
@@ -528,32 +545,9 @@ var SITE = {
   }
 };
 var README = {
-  tagline: `Landing page, API, and MCP server for discovering, shortlisting, comparing, cost-modeling, and live-testing ${MODEL_COUNT} AI models.`,
   mcpDescription: `Discover, shortlist, compare, cost-model, and live-test ${MODEL_COUNT} AI models from your editor`,
-  monorepoLayout: {
-    appsWeb: "apps/web \u2014 Next.js 16 app (UI + API routes)",
-    packagesCore: "packages/core \u2014 Shared Zod schemas, types, constants (@index9/core)",
-    packagesMcp: "packages/mcp \u2014 Thin MCP stdio server calling the hosted API (@index9/mcp)"
-  },
-  quickStart: {
-    install: "pnpm install",
-    build: "pnpm build",
-    test: "pnpm test",
-    dev: "pnpm dev    # run web app"
-  },
-  envNote: "Copy apps/web/.env.example to apps/web/.env.local and fill in values for local development.",
   mcpInstall: {
-    cli: "npx -y @index9/mcp@latest",
-    envNote: "Optional: set OPENROUTER_API_KEY in your MCP client config for live test_model calls. dryRun=true works without a key.",
-    claudeCode: "Claude Code: Run `claude mcp add --transport stdio index9 -- npx -y @index9/mcp` or add the same config to .mcp.json / ~/.claude.json."
-  },
-  release: {
-    step1: "Make changes in packages/mcp (core is internal, bundled into mcp)",
-    step2: "Run pnpm changeset \u2014 add a changeset, select packages, choose bump type",
-    step3: "Commit and push; open PR to main",
-    step4: "Merge the PR; CI creates a Version Packages PR when changesets exist",
-    step5: "Merge the version PR; CI publishes to npm and creates a GitHub Release with the .mcpb artifact attached",
-    step6: "Users can install via npx @index9/mcp@latest or download .mcpb from Releases"
+    envNote: "Optional: set OPENROUTER_API_KEY in your MCP client config for live test_model calls. dryRun=true works without a key."
   }
 };
 
@@ -762,6 +756,7 @@ import { z as z6 } from "zod";
 var ResponseFormatSchema = z6.object({
   type: z6.string().min(1)
 }).catchall(z6.unknown()).optional();
+var ProviderSortSchema = z6.enum(["throughput", "price", "latency"]);
 var TestRequestSchema = z6.object({
   prompt: z6.string().min(1).optional(),
   userContent: z6.array(UserContentPartSchema).min(1).optional(),
@@ -777,7 +772,30 @@ var TestRequestSchema = z6.object({
   seed: z6.number().int().optional(),
   responseFormat: ResponseFormatSchema,
   enforceJson: z6.boolean().optional(),
-  retries: z6.number().int().min(0).max(3).optional()
+  retries: z6.number().int().min(0).max(3).optional(),
+  // Use OpenRouter's SSE streaming endpoint so capacity/refusal errors
+  // surface in ~1s instead of waiting the full per-model timeout for an
+  // empty 200 OK. Cost/tokens are still returned via stream_options.
+  stream: z6.boolean().optional(),
+  // First-token deadline (streaming only). If the upstream sends no
+  // delta within this window, abort the request. Defaults to 10s when
+  // streaming. Ignored when stream=false.
+  firstTokenTimeoutMs: z6.number().int().positive().optional(),
+  // Forwards as `provider.sort` to OpenRouter — opt into routing toward
+  // higher-throughput providers when running benchmarks.
+  providerSort: ProviderSortSchema.optional(),
+  // Forwards as `provider.order` — try these provider slugs first in the
+  // given order before falling back. Capped to stay within reasonable
+  // limits and prevent abuse.
+  providerOrder: z6.array(z6.string().min(1)).min(1).max(8).optional(),
+  // Forwards as the top-level `models` array (NOT `model`). OpenRouter
+  // tries each in order if the primary is unavailable. Different intent
+  // from providerOrder, which routes within a single model.
+  fallbackModels: z6.array(z6.string().min(1)).min(1).max(5).optional(),
+  // When true, attach a `debug` field on each result with the raw
+  // upstream finish_reason, error message, provider name, refusal, and
+  // usage. Used to diagnose "missing assistant text" without re-running.
+  debug: z6.boolean().optional()
 }).strict().superRefine((data, ctx) => {
   if (data.dryRun === true) {
     if (!data.prompt && data.expectedPromptTokens === void 0) {
@@ -812,10 +830,28 @@ var TestFailureReasonSchema = z6.enum([
   "insufficient_credits",
   "model_unavailable",
   "rate_limited",
+  // Provider is overloaded / "at capacity" / "provisioned throughput
+  // required". A distinct reason from rate_limited so callers can apply
+  // a longer backoff or route to a fallback model.
+  "capacity",
   "timeout",
   "invalid_request",
   "unknown"
 ]);
+var TestDebugInfoSchema = z6.object({
+  upstreamId: z6.string().optional(),
+  providerName: z6.string().optional(),
+  modelPublisher: z6.string().optional(),
+  finishReason: z6.string().optional(),
+  upstreamError: z6.string().optional(),
+  refusal: z6.string().optional(),
+  hasToolCalls: z6.boolean().optional(),
+  usage: z6.object({
+    promptTokens: z6.number().optional(),
+    completionTokens: z6.number().optional(),
+    totalTokens: z6.number().optional()
+  }).optional()
+});
 var TestModelMetadataSchema = z6.object({
   id: z6.string(),
   name: z6.string(),
@@ -832,7 +868,8 @@ var TestResultSuccessSchema = z6.object({
   latencyMs: z6.number().min(0),
   tokens: UsageTokensSchema,
   cost: z6.number().nullable().optional(),
-  truncated: z6.boolean().optional()
+  truncated: z6.boolean().optional(),
+  debug: TestDebugInfoSchema.optional()
 });
 var TestResultFailureSchema = z6.object({
   modelId: z6.string(),
@@ -841,7 +878,8 @@ var TestResultFailureSchema = z6.object({
   model: TestModelMetadataSchema,
   error: z6.string(),
   failureReason: TestFailureReasonSchema.optional(),
-  latencyMs: z6.number().min(0)
+  latencyMs: z6.number().min(0),
+  debug: TestDebugInfoSchema.optional()
 });
 var TestResultSchema = z6.discriminatedUnion("ok", [
   TestResultSuccessSchema,
@@ -902,8 +940,8 @@ function loadConfig() {
 }
 
 // src/client.ts
-var RETRY_DELAYS_MS = [1e3, 2e3, 4e3];
-var ATTEMPT_TIMEOUT_MS = 3e4;
+var DEFAULT_RETRY_DELAYS_MS = [1e3, 2e3, 4e3];
+var DEFAULT_ATTEMPT_TIMEOUT_MS = 3e4;
 function isRetryable(status) {
   return status === 429 || status >= 500;
 }
@@ -919,14 +957,17 @@ function toErrorMessage(error) {
   if (error instanceof Error && error.message.trim()) return error.message;
   return "Unknown error";
 }
-async function fetchWithRetry(url, options) {
+async function fetchWithRetry(url, options, retryOptions) {
+  const attemptTimeoutMs = retryOptions?.attemptTimeoutMs ?? DEFAULT_ATTEMPT_TIMEOUT_MS;
+  const maxRetries = Math.max(0, retryOptions?.maxRetries ?? DEFAULT_RETRY_DELAYS_MS.length);
+  const retryDelaysMs = DEFAULT_RETRY_DELAYS_MS.slice(0, maxRetries);
   let lastResponse = null;
   let lastError;
-  for (let i = 0; i <= RETRY_DELAYS_MS.length; i++) {
+  for (let i = 0; i <= maxRetries; i++) {
     const timeoutController = new AbortController();
     const timeoutId = setTimeout(() => {
       timeoutController.abort(new DOMException("Request timed out", "AbortError"));
-    }, ATTEMPT_TIMEOUT_MS);
+    }, attemptTimeoutMs);
     const externalSignal = options.signal;
     const onAbort = () => {
       timeoutController.abort(
@@ -951,14 +992,12 @@ async function fetchWithRetry(url, options) {
       clearTimeout(timeoutId);
       externalSignal?.removeEventListener("abort", onAbort);
     }
-    if (i < RETRY_DELAYS_MS.length) {
-      await sleep(RETRY_DELAYS_MS[i]);
+    if (i < retryDelaysMs.length) {
+      await sleep(retryDelaysMs[i]);
     }
   }
   if (lastResponse) return lastResponse;
-  throw new Error(
-    `Request failed after ${RETRY_DELAYS_MS.length + 1} attempts: ${toErrorMessage(lastError)}`
-  );
+  throw new Error(`Request failed after ${maxRetries + 1} attempts: ${toErrorMessage(lastError)}`);
 }
 function buildUrl(baseUrl, path, params) {
   const url = new URL(path, baseUrl);
@@ -1037,8 +1076,8 @@ function extractRecoveryFields(body) {
   }
   return out;
 }
-async function callApi(ctx, url, options, responseSchema) {
-  const res = await fetchWithRetry(url, options);
+async function callApi(ctx, url, options, responseSchema, retryOptions) {
+  const res = await fetchWithRetry(url, options, retryOptions);
   let body;
   try {
     body = await res.json();
@@ -1153,7 +1192,12 @@ async function handleTestModels(ctx, args) {
     ctx,
     `${ctx.baseUrl}${API_PATHS.test}`,
     { method: "POST", headers: reqHeaders, body: JSON.stringify(parsed.data) },
-    TestResponseSchema
+    TestResponseSchema,
+    // Live inference is non-idempotent and slow: each retry costs real money
+    // and the server-side per-model retry/backoff already handles transient
+    // errors. Give the call enough wall-clock to cover a worst-case 10-model
+    // batch × 60s per model and let the server decide on retries.
+    { attemptTimeoutMs: 24e4, maxRetries: 0 }
   );
 }
 
@@ -1177,8 +1221,8 @@ async function createServer() {
         sortOrder: z7.enum(["asc", "desc"]).optional().describe("Sort order. Defaults by sortBy."),
         createdAfter: z7.string().optional().describe("Lower bound for model created timestamp."),
         createdBefore: z7.string().optional().describe("Upper bound for model created timestamp."),
-        minPrice: z7.number().min(0).optional().describe("Minimum prompt price in USD per million tokens."),
-        maxPrice: z7.number().min(0).optional().describe("Maximum prompt price in USD per million tokens."),
+        minPrice: z7.number().min(0).optional().describe(PARAM_DESCRIPTIONS.minPrice),
+        maxPrice: z7.number().min(0).optional().describe(PARAM_DESCRIPTIONS.maxPrice),
         minContext: z7.number().int().min(1).optional().describe("Minimum context window in tokens."),
         capabilitiesAll: z7.array(z7.enum(CAPABILITIES)).optional().describe(PARAM_DESCRIPTIONS.capabilitiesAll),
         capabilitiesAny: z7.array(z7.enum(CAPABILITIES)).optional().describe(PARAM_DESCRIPTIONS.capabilitiesAny),
@@ -1264,7 +1308,21 @@ async function createServer() {
           "Structured output shape request forwarded to OpenRouter (e.g., { type: 'json_object' })."
         ),
         enforceJson: z7.boolean().optional().describe("When true, output must parse as JSON."),
-        retries: z7.number().int().min(0).max(3).optional().describe("Retries for transient failures.")
+        retries: z7.number().int().min(0).max(3).optional().describe("Retries for transient failures."),
+        stream: z7.boolean().optional().describe(
+          "Use OpenRouter SSE streaming so capacity/refusal errors surface quickly. Defaults to false."
+        ),
+        firstTokenTimeoutMs: z7.number().int().min(1).optional().describe("Streaming-only first-token deadline in ms. Defaults to 10000."),
+        providerSort: ProviderSortSchema.optional().describe(
+          'OpenRouter provider routing sort: "throughput", "price", or "latency".'
+        ),
+        providerOrder: z7.array(z7.string().min(1)).min(1).max(8).optional().describe("Provider slugs to try first, in order. Up to 8."),
+        fallbackModels: z7.array(z7.string().min(1)).min(1).max(5).optional().describe(
+          "Fallback model IDs OpenRouter may try if the primary is unavailable. Up to 5."
+        ),
+        debug: z7.boolean().optional().describe(
+          "When true, include upstream finish_reason, provider, error, refusal, and usage."
+        )
       },
       // No outputSchema: test_model returns a z.union of dry-run and live shapes.
       // The SDK supports only ZodRawShape | AnySchema for outputSchema; a discriminated-union

package/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": "0.3",
   "name": "index9",
-  "version": "6.1.0",
+  "version": "6.3.0",
   "description": "Discover, shortlist, compare, cost-model, and live-test 300+ AI models from your editor",
   "author": {
     "name": "Index9"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@index9/mcp",
-  "version": "6.2.0",
+  "version": "6.4.0",
   "license": "MIT",
   "repository": {
     "type": "git",
@@ -24,11 +24,11 @@
     "zod": "^4.4.3"
   },
   "devDependencies": {
-    "@types/node": "^25.6.2",
+    "@types/node": "^25.8.0",
     "tsup": "^8.5.1",
     "typescript": "6.0.3",
     "vitest": "^4.1.6",
-    "@index9/core": "2.5.0"
+    "@index9/core": "2.7.0"
   },
   "engines": {
     "node": ">=20"