@blockrun/mcp 0.16.1 → 0.18.0

package/dist/index.js

@@ -14,6 +14,7 @@ import {
   ImageClient,
   PriceClient,
   SolanaLLMClient,
+  AnthropicClient,
   getOrCreateWallet,
   getOrCreateSolanaWallet,
   loadSolanaWallet,
@@ -33,6 +34,7 @@ var _priceClient = null;
 var _freePriceClient = null;
 var _evmWalletInfo = null;
 var _solanaClient = null;
+var _anthropicClient = null;
 function readChainPreference() {
   for (const file of CHAIN_PREFERENCE_FILES) {
     try {
@@ -58,6 +60,7 @@ var CHAIN_FILE = path.join(BLOCKRUN_DIR, ".chain");
 function resetChainCaches() {
   _evmClient = null;
   _solanaClient = null;
+  _anthropicClient = null;
   _imageClient = null;
   _priceClient = null;
   _freePriceClient = null;
@@ -78,6 +81,12 @@ async function ensureBothWallets() {
     solana: { address: sol.address, isNew: sol.isNew }
   };
 }
+function baseOnlyMessage(capability) {
+  if (getChain() === "solana") {
+    return `${capability} currently supports Base-chain payment only \u2014 your active chain is Solana. Switch with: blockrun_wallet action:"chain" chain:"base"  (switch back later with chain:"solana"). Solana support depends on the @blockrun/llm SDK adding this capability.`;
+  }
+  return null;
+}
 function ensureEvmWallet() {
   if (!_evmWalletInfo) {
     _evmWalletInfo = getOrCreateWallet();
@@ -108,6 +117,13 @@ function getClient() {
   }
   return _evmClient;
 }
+function getAnthropicClient() {
+  if (!_anthropicClient) {
+    const privateKey = getOrCreateWalletKey();
+    _anthropicClient = new AnthropicClient({ privateKey });
+  }
+  return _anthropicClient;
+}
 function getImageClient() {
   if (!_imageClient) {
     const privateKey = getOrCreateWalletKey();
@@ -214,11 +230,11 @@ var BASE_CHAIN_ID = "8453";
 var MODEL_TIERS = {
   fast: ["google/gemini-3.5-flash", "google/gemini-2.5-flash", "google/gemini-3.1-flash-lite", "openai/gpt-5-mini", "deepseek/deepseek-chat", "google/gemini-3-flash-preview"],
   balanced: ["openai/gpt-5.4", "anthropic/claude-sonnet-4.6", "google/gemini-2.5-pro", "moonshot/kimi-k2.6", "openai/gpt-5.3", "google/gemini-3.1-pro"],
-  powerful: ["openai/gpt-5.4-pro", "anthropic/claude-opus-4.7", "anthropic/claude-opus-4.6", "openai/o3", "openai/gpt-5.4"],
+  powerful: ["anthropic/claude-opus-4.8", "openai/gpt-5.4-pro", "anthropic/claude-opus-4.7", "anthropic/claude-opus-4.6", "openai/o3", "openai/gpt-5.4"],
   cheap: ["zai/glm-5", "zai/glm-5-turbo", "nvidia/gpt-oss-120b", "nvidia/deepseek-v3.2", "google/gemini-2.5-flash", "deepseek/deepseek-chat", "openai/gpt-5.4-nano"],
-  reasoning: ["openai/o3", "openai/o1", "openai/o3-mini", "deepseek/deepseek-reasoner", "moonshot/kimi-k2.6", "openai/gpt-5.3-codex"],
+  reasoning: ["anthropic/claude-opus-4.8", "openai/o3", "openai/o1", "openai/o3-mini", "deepseek/deepseek-reasoner", "moonshot/kimi-k2.6", "openai/gpt-5.3-codex"],
   free: ["nvidia/qwen3-next-80b-a3b-thinking", "nvidia/mistral-small-4-119b", "nvidia/gpt-oss-120b", "nvidia/deepseek-v3.2", "nvidia/qwen3-coder-480b", "nvidia/llama-4-maverick", "nvidia/gpt-oss-20b", "nvidia/glm-4.7"],
-  coding: ["zai/glm-5", "openai/gpt-5.3-codex", "moonshot/kimi-k2.6", "nvidia/qwen3-coder-480b", "anthropic/claude-sonnet-4.6", "openai/gpt-5.4"],
+  coding: ["anthropic/claude-opus-4.8", "zai/glm-5", "openai/gpt-5.3-codex", "moonshot/kimi-k2.6", "nvidia/qwen3-coder-480b", "anthropic/claude-sonnet-4.6", "openai/gpt-5.4"],
   glm: ["zai/glm-5", "zai/glm-5-turbo", "nvidia/glm-4.7"]
 };
 
@@ -669,6 +685,134 @@ function recordSpending(budget, cost, agentId) {
   }
 }
 
+// src/tools/chat-anthropic.ts
+function isAnthropicModel(model) {
+  const id = model.trim();
+  return /^anthropic\//i.test(id) || /^claude-/i.test(id);
+}
+function parseDataUri(url) {
+  const match = /^data:(image\/(?:jpeg|png|gif|webp));base64,(.+)$/i.exec(url.trim());
+  if (!match) return null;
+  return {
+    type: "base64",
+    media_type: match[1].toLowerCase(),
+    data: match[2]
+  };
+}
+function toAnthropicContent(content) {
+  if (typeof content === "string") return content;
+  const blocks = [];
+  for (const part of content) {
+    if (part.type === "text") {
+      if (part.text) blocks.push({ type: "text", text: part.text });
+    } else if (part.type === "image_url") {
+      const url = part.image_url?.url;
+      if (!url) continue;
+      const base64 = parseDataUri(url);
+      const source = base64 ? base64 : { type: "url", url };
+      blocks.push({ type: "image", source });
+    }
+  }
+  return blocks;
+}
+function isTextBlock(b) {
+  return b.type === "text";
+}
+function isThinkingBlock(b) {
+  return b.type === "thinking";
+}
+async function handleAnthropicNative(args) {
+  const {
+    client,
+    model,
+    message,
+    system,
+    messages,
+    maxTokens,
+    temperature,
+    stop,
+    thinking,
+    budget,
+    agentId,
+    estimatedCost
+  } = args;
+  const systemParts = [];
+  if (system) systemParts.push(system);
+  const apiMessages = [];
+  for (const m of messages ?? []) {
+    if (m.role === "system") {
+      const text = typeof m.content === "string" ? m.content : m.content.filter((p) => p.type === "text").map((p) => p.text).join("\n");
+      if (text) systemParts.push(text);
+      continue;
+    }
+    apiMessages.push({ role: m.role, content: toAnthropicContent(m.content) });
+  }
+  if (message.trim()) apiMessages.push({ role: "user", content: message });
+  if (apiMessages.length === 0) {
+    return { content: [{ type: "text", text: "No message content to send." }], isError: true };
+  }
+  let effectiveMax = maxTokens ?? 1024;
+  let raisedMaxTokens = false;
+  if (thinking && effectiveMax <= thinking.budget_tokens) {
+    effectiveMax = thinking.budget_tokens + 1024;
+    raisedMaxTokens = true;
+  }
+  const params = {
+    model,
+    max_tokens: effectiveMax,
+    messages: apiMessages
+  };
+  if (systemParts.length) params.system = systemParts.join("\n\n");
+  if (stop && stop.length) params.stop_sequences = stop;
+  if (thinking) {
+    params.thinking = { type: "enabled", budget_tokens: thinking.budget_tokens };
+  } else if (temperature !== void 0) {
+    params.temperature = Math.max(0, Math.min(1, temperature));
+  }
+  let native;
+  try {
+    native = await client.messages.create(params);
+  } catch (error) {
+    const errorMessage = error instanceof Error ? error.message : String(error);
+    return { content: [{ type: "text", text: formatError(errorMessage) }], isError: true };
+  }
+  recordSpending(budget, estimatedCost, agentId);
+  const thinkingBlocks = native.content.filter(isThinkingBlock);
+  const textBlocks = native.content.filter(isTextBlock);
+  const answerText = textBlocks.map((b) => b.text).join("\n");
+  const thinkingText = thinkingBlocks.map((b) => b.thinking).join("\n");
+  const signaturePresent = thinkingBlocks.some(
+    (b) => typeof b.signature === "string" && b.signature.length > 0
+  );
+  const headerBits = [native.model, "native /v1/messages", `thinking ${thinking ? "on" : "off"}`];
+  if (raisedMaxTokens) headerBits.push(`max_tokens\u2192${effectiveMax}`);
+  const header = `[${headerBits.join(" | ")}]`;
+  const content = [{ type: "text", text: `${header}
+
+${answerText}` }];
+  if (thinkingText) {
+    content.push({ type: "text", text: `\u{1F9E0} Thinking (signature ${signaturePresent ? "present" : "absent"}):
+${thinkingText}` });
+  }
+  return {
+    content,
+    structuredContent: {
+      requested_model: model,
+      // Verbatim upstream model id — proof the call hit real Claude with no
+      // substitution. Intentionally NOT rewritten back to the requested id.
+      model: native.model,
+      response: answerText,
+      thinking: thinkingText || void 0,
+      // Native thinking blocks verbatim, including their original signature.
+      thinking_blocks: thinkingBlocks,
+      signature_present: signaturePresent,
+      stop_reason: native.stop_reason,
+      usage: native.usage,
+      native
+    }
+  };
+}
+
 // src/tools/chat.ts
 function estimateChatCost(mode, model, routing, routingProfile) {
   if (mode === "free") return 0;
@@ -714,15 +858,28 @@ Run blockrun_models to see all 41+ models with pricing.`,
         system: z2.string().optional().describe("Optional system prompt"),
         max_tokens: z2.number().optional().default(1024).describe("Max tokens in response"),
         temperature: z2.number().optional().default(1).describe("Creativity 0-2"),
+        response_format: z2.enum(["text", "json_object"]).optional().describe("Set to 'json_object' to force valid JSON output (no markdown fences). Works across all providers."),
+        stop: z2.array(z2.string()).max(4).optional().describe("Up to 4 stop sequences; generation halts when any is produced"),
+        thinking: z2.object({
+          type: z2.literal("enabled"),
+          budget_tokens: z2.number().min(1).describe("Tokens Claude may spend reasoning before answering. max_tokens is auto-raised above this if needed.")
+        }).optional().describe("Anthropic extended thinking. Only honored for anthropic/claude-* models \u2014 these go direct to the native /v1/messages endpoint and the response includes verbatim type:'thinking' blocks with their original signature. Ignored for non-Claude models (no native thinking channel)."),
         agent_id: z2.string().optional().describe("Agent identifier. If a budget was delegated for this agent_id via blockrun_wallet action:'delegate', spending is tracked and enforced. The agent is hard-stopped when its budget is exhausted."),
         messages: z2.array(z2.object({
           role: z2.enum(["user", "assistant", "system"]),
-          content: z2.string()
+          content: z2.union([
+            z2.string(),
+            z2.array(z2.union([
+              z2.object({ type: z2.literal("text"), text: z2.string() }),
+              z2.object({ type: z2.literal("image_url"), image_url: z2.object({ url: z2.string().describe("https URL or data:<mime>;base64,<...> URI") }) })
+            ]))
+          ]).describe("Plain text, or an array of parts for multimodal input (text + image_url). Images are honored on the native anthropic/claude-* path.")
         })).optional().describe("Conversation history for multi-turn context. When provided, 'message' is appended as the final user turn. Use with explicit 'model' param (defaults to 'openai/gpt-5.4' if not specified). Note: if you include a role:'system' entry in messages[], do not also pass the system param to avoid duplicate system messages.")
       }
     },
-    async ({ message, model, mode, routing, routing_profile, system, max_tokens, temperature, agent_id, messages }) => {
+    async ({ message, model, mode, routing, routing_profile, system, max_tokens, temperature, response_format, stop, thinking, agent_id, messages }) => {
       const llm = getClient();
+      const responseFormat = response_format ? { type: response_format } : void 0;
       const estimatedCost = estimateChatCost(mode, model, routing, routing_profile);
       const budgetCheck = checkBudget(budget, agent_id, estimatedCost);
       if (!budgetCheck.allowed) {
@@ -731,6 +888,26 @@ Run blockrun_models to see all 41+ models with pricing.`,
           isError: true
         };
       }
+      if (model && isAnthropicModel(model)) {
+        const solanaBlock = baseOnlyMessage("Native Anthropic (claude-*) calls");
+        if (solanaBlock) {
+          return { content: [{ type: "text", text: solanaBlock }], isError: true };
+        }
+        return handleAnthropicNative({
+          client: getAnthropicClient(),
+          model,
+          message,
+          system,
+          messages,
+          maxTokens: max_tokens,
+          temperature,
+          stop,
+          thinking,
+          budget,
+          agentId: agent_id,
+          estimatedCost
+        });
+      }
       if (routing === "smart") {
         if (!(llm instanceof LLMClient2)) {
           return {
@@ -744,7 +921,12 @@ Run blockrun_models to see all 41+ models with pricing.`,
             maxTokens: max_tokens,
             maxOutputTokens: max_tokens,
             temperature,
-            routingProfile: routing_profile
+            // @blockrun/llm 2.x dropped the "free" routing profile; the gateway
+            // already routes to the most cost-effective model by default, so we
+            // omit it and let ClawRouter pick (matches the SDK upgrade path).
+            routingProfile: routing_profile === "free" ? void 0 : routing_profile,
+            responseFormat,
+            stop
           });
           recordSpending(budget, result.routing.costEstimate || 1e-3, agent_id);
           return {
@@ -772,7 +954,9 @@ ${result.response}` }],
         try {
           const result = await llm.chatCompletion(targetModel, fullMessages, {
             maxTokens: max_tokens,
-            temperature
+            temperature,
+            responseFormat,
+            stop
           });
           const reply = result.choices?.[0]?.message?.content || "";
           recordSpending(budget, estimatedCost, agent_id);
@@ -792,7 +976,9 @@ ${reply}` }],
           const response = await llm.chat(model, message, {
             system,
             maxTokens: max_tokens,
-            temperature
+            temperature,
+            responseFormat,
+            stop
           });
           recordSpending(budget, estimatedCost, agent_id);
           return { content: [{ type: "text", text: response }] };
@@ -812,7 +998,9 @@ ${reply}` }],
           const response = await llm.chat(m, message, {
             system,
             maxTokens: max_tokens,
-            temperature
+            temperature,
+            responseFormat,
+            stop
           });
           recordSpending(budget, estimatedCost, agent_id);
           return {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@blockrun/mcp",
-  "version": "0.16.1",
+  "version": "0.18.0",
   "mcpName": "io.github.BlockRunAI/blockrun-mcp",
   "description": "BlockRun MCP Server - Give your AI agent web search, deep research, prediction markets, crypto data, X/Twitter intelligence. Paid via x402 micropayments.",
   "type": "module",
@@ -44,7 +44,8 @@
     "url": "https://github.com/blockrunai/blockrun-mcp/issues"
   },
   "dependencies": {
-    "@blockrun/llm": "^1.12.0",
+    "@anthropic-ai/sdk": "^0.39.0",
+    "@blockrun/llm": "^2.11.0",
     "@modelcontextprotocol/sdk": "^1.0.0",
     "open": "^11.0.0",
     "qrcode": "^1.5.4",