x402-proxy 0.10.2 → 0.10.3

package/CHANGELOG.md

@@ -7,6 +7,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.10.3] - 2026-04-01
+
+### Fixed
+- Anthropic Messages API (`/v1/messages`) SSE streaming - proxy now emits proper `event: {type}` framing and omits the `data: [DONE]` sentinel that breaks the Anthropic SDK parser
+- Anthropic error responses use correct format (`{type:"error",error:{type,message}}`) instead of OpenAI format for `/v1/messages` endpoints
+- MPP streaming usage tracking - token counts were always 0/0 because raw JSON payloads lacked the `data:` prefix the tracker expected
+- Anthropic usage accumulation from `message_start` (input tokens) and `message_delta` (output tokens) events instead of relying on a single final chunk
+- Anthropic thinking mode (`thinking.budget_tokens`) now extracted from request body for history tracking
+- MPP error messages and payment history now show the EVM wallet address instead of the Solana address
+
 ## [0.10.2] - 2026-04-01
 
 ### Fixed

package/dist/bin/cli.js

@@ -343,6 +343,8 @@ function createInferenceProxyRouteHandler(opts) {
 			if (typeof val === "string") headers[key] = val;
 		}
 		const isChatCompletion = pathSuffix.includes("/chat/completions");
+		const isMessagesApi = pathSuffix.includes("/messages");
+		const isLlmEndpoint = isChatCompletion || isMessagesApi;
 		let thinkingMode;
 		if (isChatCompletion && body) try {
 			const parsed = JSON.parse(body);
@@ -352,6 +354,10 @@ function createInferenceProxyRouteHandler(opts) {
 				body = JSON.stringify(parsed);
 			}
 		} catch {}
+		if (isMessagesApi && body) try {
+			const thinking = JSON.parse(body).thinking;
+			if (thinking?.type === "enabled" && thinking.budget_tokens) thinkingMode = `budget_${thinking.budget_tokens}`;
+		} catch {}
 		const method = req.method ?? "GET";
 		const startMs = Date.now();
 		try {
@@ -361,7 +367,7 @@ function createInferenceProxyRouteHandler(opts) {
 				body: ["GET", "HEAD"].includes(method) ? void 0 : body
 			};
 			const useMpp = provider.protocol === "mpp" || provider.protocol === "auto";
-			const wantsStreaming = isChatCompletion && /"stream"\s*:\s*true/.test(body);
+			const wantsStreaming = isLlmEndpoint && /"stream"\s*:\s*true/.test(body);
 			if (useMpp) {
 				const evmKey = getEvmKey();
 				if (!evmKey) {
@@ -377,12 +383,13 @@ function createInferenceProxyRouteHandler(opts) {
 					res,
 					upstreamUrl,
 					requestInit,
-					walletAddress,
+					walletAddress: getWalletAddressForNetwork?.("eip155:4217") ?? walletAddress,
 					historyPath,
 					logger,
 					allModels,
 					thinkingMode,
 					wantsStreaming,
+					isMessagesApi,
 					startMs,
 					evmKey,
 					mppSessionBudget: provider.mppSessionBudget
@@ -421,15 +428,10 @@ function createInferenceProxyRouteHandler(opts) {
 				if (responseBody.includes("simulation") || responseBody.includes("Simulation")) userMessage = `Insufficient USDC or SOL in wallet ${walletAddress}. Fund it with USDC (SPL token) to pay for inference.`;
 				else if (responseBody.includes("insufficient") || responseBody.includes("balance")) userMessage = `Insufficient funds in wallet ${walletAddress}. Top up with USDC on Solana mainnet.`;
 				else userMessage = `x402 payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`;
-				res.writeHead(402, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: userMessage,
-					type: "x402_payment_error",
-					code: "payment_failed"
-				} }));
+				writeErrorResponse(res, 402, userMessage, "x402_payment_error", "payment_failed", isMessagesApi);
 				return true;
 			}
-			if (!response.ok && isChatCompletion) {
+			if (!response.ok && isLlmEndpoint) {
 				const responseBody = await response.text();
 				logger.error(`x402: upstream error ${response.status}: ${responseBody.substring(0, 300)}`);
 				const payment = proxy.shiftPayment();
@@ -448,12 +450,7 @@ function createInferenceProxyRouteHandler(opts) {
 					ms: Date.now() - startMs,
 					error: `upstream_${response.status}`
 				});
-				res.writeHead(502, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: `LLM provider temporarily unavailable (HTTP ${response.status}). Try again shortly.`,
-					type: "x402_upstream_error",
-					code: "upstream_failed"
-				} }));
+				writeErrorResponse(res, 502, `LLM provider temporarily unavailable (HTTP ${response.status}). Try again shortly.`, "x402_upstream_error", "upstream_failed", isMessagesApi);
 				return true;
 			}
 			logger.info(`x402: response ${response.status}`);
@@ -482,7 +479,7 @@ function createInferenceProxyRouteHandler(opts) {
 				return true;
 			}
 			const ct = response.headers.get("content-type") || "";
-			const sse = isChatCompletion && ct.includes("text/event-stream") ? createSseTracker() : null;
+			const sse = isLlmEndpoint && ct.includes("text/event-stream") ? createSseTracker() : null;
 			const reader = response.body.getReader();
 			const decoder = new TextDecoder();
 			try {
@@ -496,56 +493,17 @@ function createInferenceProxyRouteHandler(opts) {
 				reader.releaseLock();
 			}
 			res.end();
-			const durationMs = Date.now() - startMs;
-			if (sse?.lastData) try {
-				const parsed = JSON.parse(sse.lastData);
-				const usage = parsed.usage;
-				const model = parsed.model ?? "";
-				appendHistory(historyPath, {
-					t: Date.now(),
-					ok: true,
-					kind: "x402_inference",
-					net: paymentNetwork,
-					from: paymentFrom,
-					to: payment?.payTo,
-					tx: txSig,
-					amount,
-					token: "USDC",
-					provider: allModels.find((m) => m.id === model || `${m.provider}/${m.id}` === model)?.provider,
-					model,
-					inputTokens: usage?.prompt_tokens ?? 0,
-					outputTokens: usage?.completion_tokens ?? 0,
-					reasoningTokens: usage?.completion_tokens_details?.reasoning_tokens,
-					cacheRead: usage?.prompt_tokens_details?.cached_tokens,
-					cacheWrite: usage?.prompt_tokens_details?.cache_creation_input_tokens,
-					thinking: thinkingMode,
-					ms: durationMs
-				});
-			} catch {
-				appendHistory(historyPath, {
-					t: Date.now(),
-					ok: true,
-					kind: "x402_inference",
-					net: paymentNetwork,
-					from: paymentFrom,
-					to: payment?.payTo,
-					tx: txSig,
-					amount,
-					token: "USDC",
-					ms: durationMs
-				});
-			}
-			else appendHistory(historyPath, {
-				t: Date.now(),
-				ok: true,
-				kind: "x402_inference",
-				net: paymentNetwork,
-				from: paymentFrom,
-				to: payment?.payTo,
+			appendInferenceHistory({
+				historyPath,
+				allModels,
+				walletAddress: paymentFrom,
+				paymentNetwork,
+				paymentTo: payment?.payTo,
 				tx: txSig,
 				amount,
-				token: "USDC",
-				ms: durationMs
+				thinkingMode,
+				usage: sse?.result,
+				durationMs: Date.now() - startMs
 			});
 			return true;
 		} catch (err) {
@@ -565,29 +523,109 @@ function createInferenceProxyRouteHandler(opts) {
 			if (msg.includes("Simulation failed") || msg.includes("simulation")) userMessage = `Insufficient USDC or SOL in wallet ${walletAddress}. Fund it with USDC and SOL to pay for inference.`;
 			else if (msg.includes("Failed to create payment")) userMessage = `x402 payment creation failed: ${msg}. Wallet: ${walletAddress}`;
 			else userMessage = `x402 request failed: ${msg}`;
-			if (!res.headersSent) {
-				res.writeHead(402, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: userMessage,
-					type: "x402_payment_error",
-					code: "payment_failed"
-				} }));
-			}
+			if (!res.headersSent) writeErrorResponse(res, 402, userMessage, "x402_payment_error", "payment_failed", isMessagesApi);
 			return true;
 		}
 	};
 }
+function writeErrorResponse(res, status, message, type, code, isAnthropicFormat) {
+	res.writeHead(status, { "Content-Type": "application/json" });
+	if (isAnthropicFormat) res.end(JSON.stringify({
+		type: "error",
+		error: {
+			type,
+			message
+		}
+	}));
+	else res.end(JSON.stringify({ error: {
+		message,
+		type,
+		code
+	} }));
+}
 function createSseTracker() {
 	let residual = "";
-	let lastDataLine = "";
+	let anthropicModel = "";
+	let anthropicInputTokens = 0;
+	let anthropicOutputTokens = 0;
+	let anthropicCacheRead;
+	let anthropicCacheWrite;
+	let lastOpenAiData = "";
+	let isAnthropic = false;
+	function processJson(json) {
+		let parsed;
+		try {
+			parsed = JSON.parse(json);
+		} catch {
+			return;
+		}
+		const type = parsed.type;
+		if (type === "message_start") {
+			isAnthropic = true;
+			const msg = parsed.message;
+			anthropicModel = msg?.model ?? "";
+			const u = msg?.usage;
+			if (u) {
+				anthropicInputTokens = u.input_tokens ?? 0;
+				anthropicCacheWrite = u.cache_creation_input_tokens ?? void 0;
+				anthropicCacheRead = u.cache_read_input_tokens ?? void 0;
+			}
+			return;
+		}
+		if (type === "message_delta") {
+			isAnthropic = true;
+			const u = parsed.usage;
+			if (u?.output_tokens != null) anthropicOutputTokens = u.output_tokens;
+			return;
+		}
+		if (type === "message") {
+			isAnthropic = true;
+			anthropicModel = parsed.model ?? "";
+			const u = parsed.usage;
+			if (u) {
+				anthropicInputTokens = u.input_tokens ?? 0;
+				anthropicOutputTokens = u.output_tokens ?? 0;
+				anthropicCacheWrite = u.cache_creation_input_tokens ?? void 0;
+				anthropicCacheRead = u.cache_read_input_tokens ?? void 0;
+			}
+			return;
+		}
+		if (parsed.usage || parsed.model) lastOpenAiData = json;
+	}
 	return {
 		push(text) {
 			const lines = (residual + text).split("\n");
 			residual = lines.pop() ?? "";
-			for (const line of lines) if (line.startsWith("data: ") && line !== "data: [DONE]") lastDataLine = line.slice(6);
+			for (const line of lines) if (line.startsWith("data: ") && line !== "data: [DONE]") processJson(line.slice(6));
+		},
+		pushJson(text) {
+			processJson(text);
 		},
-		get lastData() {
-			return lastDataLine;
+		get result() {
+			if (isAnthropic) {
+				if (!anthropicModel && !anthropicInputTokens && !anthropicOutputTokens) return void 0;
+				return {
+					model: anthropicModel,
+					inputTokens: anthropicInputTokens,
+					outputTokens: anthropicOutputTokens,
+					cacheRead: anthropicCacheRead,
+					cacheWrite: anthropicCacheWrite
+				};
+			}
+			if (!lastOpenAiData) return void 0;
+			try {
+				const parsed = JSON.parse(lastOpenAiData);
+				return {
+					model: parsed.model ?? "",
+					inputTokens: parsed.usage?.prompt_tokens ?? 0,
+					outputTokens: parsed.usage?.completion_tokens ?? 0,
+					reasoningTokens: parsed.usage?.completion_tokens_details?.reasoning_tokens,
+					cacheRead: parsed.usage?.prompt_tokens_details?.cached_tokens,
+					cacheWrite: parsed.usage?.prompt_tokens_details?.cache_creation_input_tokens
+				};
+			} catch {
+				return;
+			}
 		}
 	};
 }
@@ -597,7 +635,7 @@ async function closeMppSession(handler) {
 	} catch {}
 }
 async function handleMppRequest(opts) {
-	const { req, res, upstreamUrl, requestInit, walletAddress, historyPath, logger, allModels, thinkingMode, wantsStreaming, startMs, evmKey, mppSessionBudget } = opts;
+	const { req, res, upstreamUrl, requestInit, walletAddress, historyPath, logger, allModels, thinkingMode, wantsStreaming, isMessagesApi, startMs, evmKey, mppSessionBudget } = opts;
 	const mpp = await createMppProxyHandler({
 		evmKey,
 		maxDeposit: mppSessionBudget
@@ -611,10 +649,22 @@ async function handleMppRequest(opts) {
 			});
 			const sse = createSseTracker();
 			const stream = await mpp.sse(upstreamUrl, requestInit);
-			for await (const chunk of stream) {
+			if (isMessagesApi) for await (const chunk of stream) {
 				const text = String(chunk);
-				res.write(text);
-				sse.push(text);
+				let eventType = "unknown";
+				try {
+					eventType = JSON.parse(text).type ?? "unknown";
+				} catch {}
+				res.write(`event: ${eventType}\ndata: ${text}\n\n`);
+				sse.pushJson(text);
+			}
+			else {
+				for await (const chunk of stream) {
+					const text = String(chunk);
+					res.write(`data: ${text}\n\n`);
+					sse.pushJson(text);
+				}
+				res.write("data: [DONE]\n\n");
 			}
 			res.end();
 			mpp.shiftPayment();
@@ -628,7 +678,7 @@ async function handleMppRequest(opts) {
 				tx: payment?.receipt?.reference ?? payment?.channelId,
 				amount: parseMppAmount(payment?.amount),
 				thinkingMode,
-				lastDataLine: sse.lastData,
+				usage: sse.result,
 				durationMs: Date.now() - startMs
 			});
 			return true;
@@ -647,12 +697,7 @@ async function handleMppRequest(opts) {
 				ms: Date.now() - startMs,
 				error: "payment_required"
 			});
-			res.writeHead(402, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({ error: {
-				message: `MPP payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`,
-				type: "mpp_payment_error",
-				code: "payment_failed"
-			} }));
+			writeErrorResponse(res, 402, `MPP payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`, "mpp_payment_error", "payment_failed", isMessagesApi);
 			return true;
 		}
 		const resHeaders = {};
@@ -660,6 +705,8 @@ async function handleMppRequest(opts) {
 		res.writeHead(response.status, resHeaders);
 		const responseBody = await response.text();
 		res.end(responseBody);
+		const usageTracker = createSseTracker();
+		usageTracker.pushJson(responseBody);
 		const payment = mpp.shiftPayment();
 		appendInferenceHistory({
 			historyPath,
@@ -670,7 +717,7 @@ async function handleMppRequest(opts) {
 			tx: tx ?? payment?.receipt?.reference,
 			amount: parseMppAmount(payment?.amount),
 			thinkingMode,
-			lastDataLine: responseBody,
+			usage: usageTracker.result,
 			durationMs: Date.now() - startMs
 		});
 		return true;
@@ -685,14 +732,7 @@ async function handleMppRequest(opts) {
 			ms: Date.now() - startMs,
 			error: String(err).substring(0, 200)
 		});
-		if (!res.headersSent) {
-			res.writeHead(402, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({ error: {
-				message: `MPP request failed: ${String(err)}`,
-				type: "mpp_payment_error",
-				code: "payment_failed"
-			} }));
-		}
+		if (!res.headersSent) writeErrorResponse(res, 402, `MPP request failed: ${String(err)}`, "mpp_payment_error", "payment_failed", isMessagesApi);
 		return true;
 	} finally {
 		await closeMppSession(mpp);
@@ -701,45 +741,39 @@ async function handleMppRequest(opts) {
 	}
 }
 function appendInferenceHistory(opts) {
-	const { historyPath, allModels, walletAddress, paymentNetwork, paymentTo, tx, amount, thinkingMode, lastDataLine, durationMs } = opts;
-	try {
-		const parsed = JSON.parse(lastDataLine);
-		const usage = parsed.usage;
-		const model = parsed.model ?? "";
-		appendHistory(historyPath, {
-			t: Date.now(),
-			ok: true,
-			kind: "x402_inference",
-			net: paymentNetwork,
-			from: walletAddress,
-			to: paymentTo,
-			tx,
-			amount,
-			token: "USDC",
-			provider: allModels.find((entry) => entry.id === model || `${entry.provider}/${entry.id}` === model)?.provider,
-			model,
-			inputTokens: usage?.prompt_tokens ?? 0,
-			outputTokens: usage?.completion_tokens ?? 0,
-			reasoningTokens: usage?.completion_tokens_details?.reasoning_tokens,
-			cacheRead: usage?.prompt_tokens_details?.cached_tokens,
-			cacheWrite: usage?.prompt_tokens_details?.cache_creation_input_tokens,
-			thinking: thinkingMode,
-			ms: durationMs
-		});
-	} catch {
-		appendHistory(historyPath, {
-			t: Date.now(),
-			ok: true,
-			kind: "x402_inference",
-			net: paymentNetwork,
-			from: walletAddress,
-			to: paymentTo,
-			tx,
-			amount,
-			token: "USDC",
-			ms: durationMs
-		});
-	}
+	const { historyPath, allModels, walletAddress, paymentNetwork, paymentTo, tx, amount, thinkingMode, usage, durationMs } = opts;
+	if (usage) appendHistory(historyPath, {
+		t: Date.now(),
+		ok: true,
+		kind: "x402_inference",
+		net: paymentNetwork,
+		from: walletAddress,
+		to: paymentTo,
+		tx,
+		amount,
+		token: "USDC",
+		provider: allModels.find((entry) => entry.id === usage.model || `${entry.provider}/${entry.id}` === usage.model)?.provider,
+		model: usage.model,
+		inputTokens: usage.inputTokens,
+		outputTokens: usage.outputTokens,
+		reasoningTokens: usage.reasoningTokens,
+		cacheRead: usage.cacheRead,
+		cacheWrite: usage.cacheWrite,
+		thinking: thinkingMode,
+		ms: durationMs
+	});
+	else appendHistory(historyPath, {
+		t: Date.now(),
+		ok: true,
+		kind: "x402_inference",
+		net: paymentNetwork,
+		from: walletAddress,
+		to: paymentTo,
+		tx,
+		amount,
+		token: "USDC",
+		ms: durationMs
+	});
 }
 //#endregion
 //#region src/commands/serve.ts
@@ -1778,7 +1812,7 @@ Wallet is auto-generated on first run. No env vars needed.`
 			}
 			const remoteClient = new Client({
 				name: "x402-proxy",
-				version: "0.10.2"
+				version: "0.10.3"
 			});
 			const x402Mcp = new x402MCPClient(remoteClient, x402PaymentClient, {
 				autoPayment: true,
@@ -1816,7 +1850,7 @@ Wallet is auto-generated on first run. No env vars needed.`
 			}
 			const localServer = new Server({
 				name: "x402-proxy",
-				version: "0.10.2"
+				version: "0.10.3"
 			}, { capabilities: {
 				tools: tools.length > 0 ? {} : void 0,
 				resources: remoteResources.length > 0 ? {} : void 0
@@ -1911,7 +1945,7 @@ Wallet is auto-generated on first run. No env vars needed.`
 			}));
 			const remoteClient = new Client({
 				name: "x402-proxy",
-				version: "0.10.2"
+				version: "0.10.3"
 			});
 			await connectTransport(remoteClient);
 			const mppClient = McpClient.wrap(remoteClient, { methods: wrappedMethods });
@@ -1926,7 +1960,7 @@ Wallet is auto-generated on first run. No env vars needed.`
 			}
 			const localServer = new Server({
 				name: "x402-proxy",
-				version: "0.10.2"
+				version: "0.10.3"
 			}, { capabilities: {
 				tools: tools.length > 0 ? {} : void 0,
 				resources: remoteResources.length > 0 ? {} : void 0
@@ -2325,7 +2359,7 @@ const app = buildApplication(buildRouteMap({
 	docs: { brief: "curl for x402 paid APIs" }
 }), {
 	name: "x402-proxy",
-	versionInfo: { currentVersion: "0.10.2" },
+	versionInfo: { currentVersion: "0.10.3" },
 	scanner: { caseStyle: "allow-kebab-for-camel" }
 });
 //#endregion

package/dist/openclaw/plugin.js

@@ -1150,7 +1150,7 @@ function createWalletCommand(ctx) {
 			if (parts[0]?.toLowerCase() === "send") return { text: "Use `/x_send <amount|all> <address>` for transfers." };
 			try {
 				const snap = await getWalletSnapshot(ctx.rpcUrl, solanaWallet, evmWallet, ctx.historyPath);
-				const lines = [`x402-proxy v0.10.2`];
+				const lines = [`x402-proxy v0.10.3`];
 				const defaultModel = ctx.allModels[0];
 				if (defaultModel) lines.push("", `**Model** - ${defaultModel.name} (${defaultModel.provider})`);
 				lines.push("", `**Protocol** - ${ctx.getDefaultRequestProtocol()}`);
@@ -1411,6 +1411,8 @@ function createInferenceProxyRouteHandler(opts) {
 			if (typeof val === "string") headers[key] = val;
 		}
 		const isChatCompletion = pathSuffix.includes("/chat/completions");
+		const isMessagesApi = pathSuffix.includes("/messages");
+		const isLlmEndpoint = isChatCompletion || isMessagesApi;
 		let thinkingMode;
 		if (isChatCompletion && body) try {
 			const parsed = JSON.parse(body);
@@ -1420,6 +1422,10 @@ function createInferenceProxyRouteHandler(opts) {
 				body = JSON.stringify(parsed);
 			}
 		} catch {}
+		if (isMessagesApi && body) try {
+			const thinking = JSON.parse(body).thinking;
+			if (thinking?.type === "enabled" && thinking.budget_tokens) thinkingMode = `budget_${thinking.budget_tokens}`;
+		} catch {}
 		const method = req.method ?? "GET";
 		const startMs = Date.now();
 		try {
@@ -1429,7 +1435,7 @@ function createInferenceProxyRouteHandler(opts) {
 				body: ["GET", "HEAD"].includes(method) ? void 0 : body
 			};
 			const useMpp = provider.protocol === "mpp" || provider.protocol === "auto";
-			const wantsStreaming = isChatCompletion && /"stream"\s*:\s*true/.test(body);
+			const wantsStreaming = isLlmEndpoint && /"stream"\s*:\s*true/.test(body);
 			if (useMpp) {
 				const evmKey = getEvmKey();
 				if (!evmKey) {
@@ -1445,12 +1451,13 @@ function createInferenceProxyRouteHandler(opts) {
 					res,
 					upstreamUrl,
 					requestInit,
-					walletAddress,
+					walletAddress: getWalletAddressForNetwork?.("eip155:4217") ?? walletAddress,
 					historyPath,
 					logger,
 					allModels,
 					thinkingMode,
 					wantsStreaming,
+					isMessagesApi,
 					startMs,
 					evmKey,
 					mppSessionBudget: provider.mppSessionBudget
@@ -1489,15 +1496,10 @@ function createInferenceProxyRouteHandler(opts) {
 				if (responseBody.includes("simulation") || responseBody.includes("Simulation")) userMessage = `Insufficient USDC or SOL in wallet ${walletAddress}. Fund it with USDC (SPL token) to pay for inference.`;
 				else if (responseBody.includes("insufficient") || responseBody.includes("balance")) userMessage = `Insufficient funds in wallet ${walletAddress}. Top up with USDC on Solana mainnet.`;
 				else userMessage = `x402 payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`;
-				res.writeHead(402, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: userMessage,
-					type: "x402_payment_error",
-					code: "payment_failed"
-				} }));
+				writeErrorResponse(res, 402, userMessage, "x402_payment_error", "payment_failed", isMessagesApi);
 				return true;
 			}
-			if (!response.ok && isChatCompletion) {
+			if (!response.ok && isLlmEndpoint) {
 				const responseBody = await response.text();
 				logger.error(`x402: upstream error ${response.status}: ${responseBody.substring(0, 300)}`);
 				const payment = proxy.shiftPayment();
@@ -1516,12 +1518,7 @@ function createInferenceProxyRouteHandler(opts) {
 					ms: Date.now() - startMs,
 					error: `upstream_${response.status}`
 				});
-				res.writeHead(502, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: `LLM provider temporarily unavailable (HTTP ${response.status}). Try again shortly.`,
-					type: "x402_upstream_error",
-					code: "upstream_failed"
-				} }));
+				writeErrorResponse(res, 502, `LLM provider temporarily unavailable (HTTP ${response.status}). Try again shortly.`, "x402_upstream_error", "upstream_failed", isMessagesApi);
 				return true;
 			}
 			logger.info(`x402: response ${response.status}`);
@@ -1550,7 +1547,7 @@ function createInferenceProxyRouteHandler(opts) {
 				return true;
 			}
 			const ct = response.headers.get("content-type") || "";
-			const sse = isChatCompletion && ct.includes("text/event-stream") ? createSseTracker() : null;
+			const sse = isLlmEndpoint && ct.includes("text/event-stream") ? createSseTracker() : null;
 			const reader = response.body.getReader();
 			const decoder = new TextDecoder();
 			try {
@@ -1564,56 +1561,17 @@ function createInferenceProxyRouteHandler(opts) {
 				reader.releaseLock();
 			}
 			res.end();
-			const durationMs = Date.now() - startMs;
-			if (sse?.lastData) try {
-				const parsed = JSON.parse(sse.lastData);
-				const usage = parsed.usage;
-				const model = parsed.model ?? "";
-				appendHistory(historyPath, {
-					t: Date.now(),
-					ok: true,
-					kind: "x402_inference",
-					net: paymentNetwork,
-					from: paymentFrom,
-					to: payment?.payTo,
-					tx: txSig,
-					amount,
-					token: "USDC",
-					provider: allModels.find((m) => m.id === model || `${m.provider}/${m.id}` === model)?.provider,
-					model,
-					inputTokens: usage?.prompt_tokens ?? 0,
-					outputTokens: usage?.completion_tokens ?? 0,
-					reasoningTokens: usage?.completion_tokens_details?.reasoning_tokens,
-					cacheRead: usage?.prompt_tokens_details?.cached_tokens,
-					cacheWrite: usage?.prompt_tokens_details?.cache_creation_input_tokens,
-					thinking: thinkingMode,
-					ms: durationMs
-				});
-			} catch {
-				appendHistory(historyPath, {
-					t: Date.now(),
-					ok: true,
-					kind: "x402_inference",
-					net: paymentNetwork,
-					from: paymentFrom,
-					to: payment?.payTo,
-					tx: txSig,
-					amount,
-					token: "USDC",
-					ms: durationMs
-				});
-			}
-			else appendHistory(historyPath, {
-				t: Date.now(),
-				ok: true,
-				kind: "x402_inference",
-				net: paymentNetwork,
-				from: paymentFrom,
-				to: payment?.payTo,
+			appendInferenceHistory({
+				historyPath,
+				allModels,
+				walletAddress: paymentFrom,
+				paymentNetwork,
+				paymentTo: payment?.payTo,
 				tx: txSig,
 				amount,
-				token: "USDC",
-				ms: durationMs
+				thinkingMode,
+				usage: sse?.result,
+				durationMs: Date.now() - startMs
 			});
 			return true;
 		} catch (err) {
@@ -1633,29 +1591,109 @@ function createInferenceProxyRouteHandler(opts) {
 			if (msg.includes("Simulation failed") || msg.includes("simulation")) userMessage = `Insufficient USDC or SOL in wallet ${walletAddress}. Fund it with USDC and SOL to pay for inference.`;
 			else if (msg.includes("Failed to create payment")) userMessage = `x402 payment creation failed: ${msg}. Wallet: ${walletAddress}`;
 			else userMessage = `x402 request failed: ${msg}`;
-			if (!res.headersSent) {
-				res.writeHead(402, { "Content-Type": "application/json" });
-				res.end(JSON.stringify({ error: {
-					message: userMessage,
-					type: "x402_payment_error",
-					code: "payment_failed"
-				} }));
-			}
+			if (!res.headersSent) writeErrorResponse(res, 402, userMessage, "x402_payment_error", "payment_failed", isMessagesApi);
 			return true;
 		}
 	};
 }
+function writeErrorResponse(res, status, message, type, code, isAnthropicFormat) {
+	res.writeHead(status, { "Content-Type": "application/json" });
+	if (isAnthropicFormat) res.end(JSON.stringify({
+		type: "error",
+		error: {
+			type,
+			message
+		}
+	}));
+	else res.end(JSON.stringify({ error: {
+		message,
+		type,
+		code
+	} }));
+}
 function createSseTracker() {
 	let residual = "";
-	let lastDataLine = "";
+	let anthropicModel = "";
+	let anthropicInputTokens = 0;
+	let anthropicOutputTokens = 0;
+	let anthropicCacheRead;
+	let anthropicCacheWrite;
+	let lastOpenAiData = "";
+	let isAnthropic = false;
+	function processJson(json) {
+		let parsed;
+		try {
+			parsed = JSON.parse(json);
+		} catch {
+			return;
+		}
+		const type = parsed.type;
+		if (type === "message_start") {
+			isAnthropic = true;
+			const msg = parsed.message;
+			anthropicModel = msg?.model ?? "";
+			const u = msg?.usage;
+			if (u) {
+				anthropicInputTokens = u.input_tokens ?? 0;
+				anthropicCacheWrite = u.cache_creation_input_tokens ?? void 0;
+				anthropicCacheRead = u.cache_read_input_tokens ?? void 0;
+			}
+			return;
+		}
+		if (type === "message_delta") {
+			isAnthropic = true;
+			const u = parsed.usage;
+			if (u?.output_tokens != null) anthropicOutputTokens = u.output_tokens;
+			return;
+		}
+		if (type === "message") {
+			isAnthropic = true;
+			anthropicModel = parsed.model ?? "";
+			const u = parsed.usage;
+			if (u) {
+				anthropicInputTokens = u.input_tokens ?? 0;
+				anthropicOutputTokens = u.output_tokens ?? 0;
+				anthropicCacheWrite = u.cache_creation_input_tokens ?? void 0;
+				anthropicCacheRead = u.cache_read_input_tokens ?? void 0;
+			}
+			return;
+		}
+		if (parsed.usage || parsed.model) lastOpenAiData = json;
+	}
 	return {
 		push(text) {
 			const lines = (residual + text).split("\n");
 			residual = lines.pop() ?? "";
-			for (const line of lines) if (line.startsWith("data: ") && line !== "data: [DONE]") lastDataLine = line.slice(6);
+			for (const line of lines) if (line.startsWith("data: ") && line !== "data: [DONE]") processJson(line.slice(6));
+		},
+		pushJson(text) {
+			processJson(text);
 		},
-		get lastData() {
-			return lastDataLine;
+		get result() {
+			if (isAnthropic) {
+				if (!anthropicModel && !anthropicInputTokens && !anthropicOutputTokens) return void 0;
+				return {
+					model: anthropicModel,
+					inputTokens: anthropicInputTokens,
+					outputTokens: anthropicOutputTokens,
+					cacheRead: anthropicCacheRead,
+					cacheWrite: anthropicCacheWrite
+				};
+			}
+			if (!lastOpenAiData) return void 0;
+			try {
+				const parsed = JSON.parse(lastOpenAiData);
+				return {
+					model: parsed.model ?? "",
+					inputTokens: parsed.usage?.prompt_tokens ?? 0,
+					outputTokens: parsed.usage?.completion_tokens ?? 0,
+					reasoningTokens: parsed.usage?.completion_tokens_details?.reasoning_tokens,
+					cacheRead: parsed.usage?.prompt_tokens_details?.cached_tokens,
+					cacheWrite: parsed.usage?.prompt_tokens_details?.cache_creation_input_tokens
+				};
+			} catch {
+				return;
+			}
 		}
 	};
 }
@@ -1665,7 +1703,7 @@ async function closeMppSession(handler) {
 	} catch {}
 }
 async function handleMppRequest(opts) {
-	const { req, res, upstreamUrl, requestInit, walletAddress, historyPath, logger, allModels, thinkingMode, wantsStreaming, startMs, evmKey, mppSessionBudget } = opts;
+	const { req, res, upstreamUrl, requestInit, walletAddress, historyPath, logger, allModels, thinkingMode, wantsStreaming, isMessagesApi, startMs, evmKey, mppSessionBudget } = opts;
 	const mpp = await createMppProxyHandler({
 		evmKey,
 		maxDeposit: mppSessionBudget
@@ -1679,10 +1717,22 @@ async function handleMppRequest(opts) {
 			});
 			const sse = createSseTracker();
 			const stream = await mpp.sse(upstreamUrl, requestInit);
-			for await (const chunk of stream) {
+			if (isMessagesApi) for await (const chunk of stream) {
 				const text = String(chunk);
-				res.write(text);
-				sse.push(text);
+				let eventType = "unknown";
+				try {
+					eventType = JSON.parse(text).type ?? "unknown";
+				} catch {}
+				res.write(`event: ${eventType}\ndata: ${text}\n\n`);
+				sse.pushJson(text);
+			}
+			else {
+				for await (const chunk of stream) {
+					const text = String(chunk);
+					res.write(`data: ${text}\n\n`);
+					sse.pushJson(text);
+				}
+				res.write("data: [DONE]\n\n");
 			}
 			res.end();
 			mpp.shiftPayment();
@@ -1696,7 +1746,7 @@ async function handleMppRequest(opts) {
 				tx: payment?.receipt?.reference ?? payment?.channelId,
 				amount: parseMppAmount(payment?.amount),
 				thinkingMode,
-				lastDataLine: sse.lastData,
+				usage: sse.result,
 				durationMs: Date.now() - startMs
 			});
 			return true;
@@ -1715,12 +1765,7 @@ async function handleMppRequest(opts) {
 				ms: Date.now() - startMs,
 				error: "payment_required"
 			});
-			res.writeHead(402, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({ error: {
-				message: `MPP payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`,
-				type: "mpp_payment_error",
-				code: "payment_failed"
-			} }));
+			writeErrorResponse(res, 402, `MPP payment failed: ${responseBody.substring(0, 200) || "unknown error"}. Wallet: ${walletAddress}`, "mpp_payment_error", "payment_failed", isMessagesApi);
 			return true;
 		}
 		const resHeaders = {};
@@ -1728,6 +1773,8 @@ async function handleMppRequest(opts) {
 		res.writeHead(response.status, resHeaders);
 		const responseBody = await response.text();
 		res.end(responseBody);
+		const usageTracker = createSseTracker();
+		usageTracker.pushJson(responseBody);
 		const payment = mpp.shiftPayment();
 		appendInferenceHistory({
 			historyPath,
@@ -1738,7 +1785,7 @@ async function handleMppRequest(opts) {
 			tx: tx ?? payment?.receipt?.reference,
 			amount: parseMppAmount(payment?.amount),
 			thinkingMode,
-			lastDataLine: responseBody,
+			usage: usageTracker.result,
 			durationMs: Date.now() - startMs
 		});
 		return true;
@@ -1753,14 +1800,7 @@ async function handleMppRequest(opts) {
 			ms: Date.now() - startMs,
 			error: String(err).substring(0, 200)
 		});
-		if (!res.headersSent) {
-			res.writeHead(402, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({ error: {
-				message: `MPP request failed: ${String(err)}`,
-				type: "mpp_payment_error",
-				code: "payment_failed"
-			} }));
-		}
+		if (!res.headersSent) writeErrorResponse(res, 402, `MPP request failed: ${String(err)}`, "mpp_payment_error", "payment_failed", isMessagesApi);
 		return true;
 	} finally {
 		await closeMppSession(mpp);
@@ -1769,45 +1809,39 @@ async function handleMppRequest(opts) {
 	}
 }
 function appendInferenceHistory(opts) {
-	const { historyPath, allModels, walletAddress, paymentNetwork, paymentTo, tx, amount, thinkingMode, lastDataLine, durationMs } = opts;
-	try {
-		const parsed = JSON.parse(lastDataLine);
-		const usage = parsed.usage;
-		const model = parsed.model ?? "";
-		appendHistory(historyPath, {
-			t: Date.now(),
-			ok: true,
-			kind: "x402_inference",
-			net: paymentNetwork,
-			from: walletAddress,
-			to: paymentTo,
-			tx,
-			amount,
-			token: "USDC",
-			provider: allModels.find((entry) => entry.id === model || `${entry.provider}/${entry.id}` === model)?.provider,
-			model,
-			inputTokens: usage?.prompt_tokens ?? 0,
-			outputTokens: usage?.completion_tokens ?? 0,
-			reasoningTokens: usage?.completion_tokens_details?.reasoning_tokens,
-			cacheRead: usage?.prompt_tokens_details?.cached_tokens,
-			cacheWrite: usage?.prompt_tokens_details?.cache_creation_input_tokens,
-			thinking: thinkingMode,
-			ms: durationMs
-		});
-	} catch {
-		appendHistory(historyPath, {
-			t: Date.now(),
-			ok: true,
-			kind: "x402_inference",
-			net: paymentNetwork,
-			from: walletAddress,
-			to: paymentTo,
-			tx,
-			amount,
-			token: "USDC",
-			ms: durationMs
-		});
-	}
+	const { historyPath, allModels, walletAddress, paymentNetwork, paymentTo, tx, amount, thinkingMode, usage, durationMs } = opts;
+	if (usage) appendHistory(historyPath, {
+		t: Date.now(),
+		ok: true,
+		kind: "x402_inference",
+		net: paymentNetwork,
+		from: walletAddress,
+		to: paymentTo,
+		tx,
+		amount,
+		token: "USDC",
+		provider: allModels.find((entry) => entry.id === usage.model || `${entry.provider}/${entry.id}` === usage.model)?.provider,
+		model: usage.model,
+		inputTokens: usage.inputTokens,
+		outputTokens: usage.outputTokens,
+		reasoningTokens: usage.reasoningTokens,
+		cacheRead: usage.cacheRead,
+		cacheWrite: usage.cacheWrite,
+		thinking: thinkingMode,
+		ms: durationMs
+	});
+	else appendHistory(historyPath, {
+		t: Date.now(),
+		ok: true,
+		kind: "x402_inference",
+		net: paymentNetwork,
+		from: walletAddress,
+		to: paymentTo,
+		tx,
+		amount,
+		token: "USDC",
+		ms: durationMs
+	});
 }
 //#endregion
 //#region src/openclaw/plugin.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "x402-proxy",
-  "version": "0.10.2",
+  "version": "0.10.3",
   "description": "curl for x402 paid APIs. Auto-pays any endpoint on Base, Solana, and Tempo. Also works as an OpenClaw plugin.",
   "type": "module",
   "sideEffects": false,