@nqminds/mcp-client 1.0.11 → 1.0.14

package/dist/openai-client.d.ts

@@ -74,6 +74,12 @@ export declare class MCPClientOpenAI {
      * Latest usage snapshot for logging/inspection.
      */
     private lastUsage;
+    /**
+     * Per-turn accumulator — reset at the start of each processQuery / processRawQuery call.
+     * Tracks the max input tokens seen (largest context = most representative) and
+     * the sum of output tokens across all API calls in the turn.
+     */
+    private turnStats;
     private config;
     constructor(config: MCPClientConfig);
     connect(): Promise<void>;
@@ -101,7 +107,17 @@ export declare class MCPClientOpenAI {
      */
     private roughEstimateInputTokens;
     /**
-     * Normalize usage from Responses API.
+     * Reset the per-turn accumulator at the start of each processQuery / processRawQuery.
+     */
+    private startTurn;
+    /**
+     * Normalize usage from Responses API and accumulate into the per-turn stats.
+     *
+     * With previous_response_id chaining there are multiple API calls per user turn:
+     *   - inputTokens: use max across calls (the call with the largest context is most representative)
+     *   - cachedTokens: taken from the same call that has the max input tokens (keeps % ≤ 100%)
+     *   - outputTokens: sum across calls (each call generates separately-billed tokens)
+     *   - reasoningTokens: sum across calls
      */
     private captureUsage;
     /**

package/dist/openai-client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"openai-client.d.ts","sourceRoot":"","sources":["../src/openai-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAMH,MAAM,WAAW,eAAe;IAC9B,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IAEvB;;;;OAIG;IACH,yBAAyB,CAAC,EAAE,MAAM,CAAC;IAEnC;;;;OAIG;IACH,2BAA2B,CAAC,EAAE,MAAM,CAAC;IAErC;;OAEG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAID,UAAU,UAAU;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,EAAE,MAAM,CAAC;CACxB;AAiBD,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,SAAS,CAAuB;IAExC;;;OAGG;IACH,OAAO,CAAC,YAAY,CAAuB;IAE3C;;;OAGG;IACH,OAAO,CAAC,mBAAmB,CAA2B;IAEtD;;;OAGG;IACH,OAAO,CAAC,UAAU,CAGhB;IAEF;;OAEG;IACH,OAAO,CAAC,eAAe,CAAK;IAE5B;;OAEG;IACH,OAAO,CAAC,SAAS,CAOf;IAEF,OAAO,CAAC,MAAM,CAA4B;gBAE9B,MAAM,EAAE,eAAe;IAoC7B,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAIxB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAI9B,YAAY,IAAI,IAAI;IAiBpB,QAAQ,IAAI,UAAU;IAItB;;;OAGG;YACW,kBAAkB;IA0BhC;;;OAGG;IACH,OAAO,CAAC,UAAU;IAalB;;;OAGG;YACW,gBAAgB;IAuB9B;;;OAGG;IACH,OAAO,CAAC,wBAAwB;IAUhC;;OAEG;IACH,OAAO,CAAC,YAAY;IAoBpB;;OAEG;IACH,OAAO,CAAC,iBAAiB;IA8EzB,OAAO,CAAC,eAAe;IAQvB,OAAO,CAAC,kBAAkB;IAQ1B;;;;;;OAMG;IACH,OAAO,CAAC,iBAAiB;IA2CzB;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAsC9B;;OAEG;YACW,kBAAkB;IA4ChC;;;OAGG;YACW,mBAAmB;IAoBjC;;;OAGG;IACH,OAAO,CAAC,8BAA8B;IAatC;;OAEG;YACW,UAAU;IAiBxB;;OAEG;YACW,cAAc;IAoB5B;;OAEG;IACG,YAAY,CAChB,KAAK,EAAE,MAAM,EACb,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,EACtC,WAAW,CAAC,EAAE,WAAW,EACzB,kBAAkB,UAAQ,GACzB,OAAO,CAAC,MAAM,CAAC;IA6OlB;;OAEG;YACW,eAAe;CAwG9B"}
+{"version":3,"file":"openai-client.d.ts","sourceRoot":"","sources":["../src/openai-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAMH,MAAM,WAAW,eAAe;IAC9B,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IAEvB;;;;OAIG;IACH,yBAAyB,CAAC,EAAE,MAAM,CAAC;IAEnC;;;;OAIG;IACH,2BAA2B,CAAC,EAAE,MAAM,CAAC;IAErC;;OAEG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAID,UAAU,UAAU;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,EAAE,MAAM,CAAC;CACxB;AAiBD,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,SAAS,CAAuB;IAExC;;;OAGG;IACH,OAAO,CAAC,YAAY,CAAuB;IAE3C;;;OAGG;IACH,OAAO,CAAC,mBAAmB,CAA2B;IAEtD;;;OAGG;IACH,OAAO,CAAC,UAAU,CAGhB;IAEF;;OAEG;IACH,OAAO,CAAC,eAAe,CAAK;IAE5B;;OAEG;IACH,OAAO,CAAC,SAAS,CAOf;IAEF;;;;OAIG;IACH,OAAO,CAAC,SAAS,CAKf;IAEF,OAAO,CAAC,MAAM,CAA4B;gBAE9B,MAAM,EAAE,eAAe;IAoC7B,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAIxB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAI9B,YAAY,IAAI,IAAI;IAkBpB,QAAQ,IAAI,UAAU;IAItB;;;OAGG;YACW,kBAAkB;IA0BhC;;;OAGG;IACH,OAAO,CAAC,UAAU;IAalB;;;OAGG;YACW,gBAAgB;IAuB9B;;;OAGG;IACH,OAAO,CAAC,wBAAwB;IAUhC;;OAEG;IACH,OAAO,CAAC,SAAS;IAIjB;;;;;;;;OAQG;IACH,OAAO,CAAC,YAAY;IAmCpB;;OAEG;IACH,OAAO,CAAC,iBAAiB;IA8EzB,OAAO,CAAC,eAAe;IAQvB,OAAO,CAAC,kBAAkB;IAQ1B;;;;;;OAMG;IACH,OAAO,CAAC,iBAAiB;IA2CzB;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAsC9B;;OAEG;YACW,kBAAkB;IA4ChC;;;OAGG;YACW,mBAAmB;IAoBjC;;;OAGG;IACH,OAAO,CAAC,8BAA8B;IAatC;;OAEG;YACW,UAAU;IAiBxB;;OAEG;YACW,cAAc;IAoB5B;;OAEG;IACG,YAAY,CAChB,KAAK,EAAE,MAAM,EACb,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,EACtC,WAAW,CAAC,EAAE,WAAW,EACzB,kBAAkB,UAAQ,GACzB,OAAO,CAAC,MAAM,CAAC;IA8OlB;;OAEG;YACW,eAAe;CAyG9B"}

package/dist/openai-client.js

@@ -54,6 +54,17 @@ export class MCPClientOpenAI {
             reasoningTokens: 0,
             compactedTurns: 0,
         };
+        /**
+         * Per-turn accumulator — reset at the start of each processQuery / processRawQuery call.
+         * Tracks the max input tokens seen (largest context = most representative) and
+         * the sum of output tokens across all API calls in the turn.
+         */
+        this.turnStats = {
+            inputTokens: 0,
+            outputTokens: 0,
+            cachedTokens: 0,
+            reasoningTokens: 0,
+        };
         this.config = {
             openaiApiKey: config.openaiApiKey,
             mcpServerCommand: config.mcpServerCommand,
@@ -94,6 +105,7 @@ export class MCPClientOpenAI {
             compactedTurns: 0,
         };
         this.lastInputTokens = 0;
+        this.turnStats = { inputTokens: 0, outputTokens: 0, cachedTokens: 0, reasoningTokens: 0 };
         this.lastUsage = {
             inputTokens: 0,
             outputTokens: 0,
@@ -186,7 +198,19 @@ export class MCPClientOpenAI {
         return Math.ceil(serialized.length / 4);
     }
     /**
-     * Normalize usage from Responses API.
+     * Reset the per-turn accumulator at the start of each processQuery / processRawQuery.
+     */
+    startTurn() {
+        this.turnStats = { inputTokens: 0, outputTokens: 0, cachedTokens: 0, reasoningTokens: 0 };
+    }
+    /**
+     * Normalize usage from Responses API and accumulate into the per-turn stats.
+     *
+     * With previous_response_id chaining there are multiple API calls per user turn:
+     *   - inputTokens: use max across calls (the call with the largest context is most representative)
+     *   - cachedTokens: taken from the same call that has the max input tokens (keeps % ≤ 100%)
+     *   - outputTokens: sum across calls (each call generates separately-billed tokens)
+     *   - reasoningTokens: sum across calls
      */
     captureUsage(response) {
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
@@ -195,15 +219,28 @@ export class MCPClientOpenAI {
         const inputDetails = usage?.input_tokens_details ?? {};
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
         const outputDetails = usage?.output_tokens_details ?? {};
+        const inputTokens = usage.input_tokens ?? 0;
+        const outputTokens = usage.output_tokens ?? 0;
+        const cachedTokens = inputDetails.cached_tokens ?? 0;
+        const reasoningTokens = outputDetails.reasoning_tokens ?? 0;
+        // cachedTokens is only meaningful relative to its own call's inputTokens.
+        // Track it alongside the max-input call so the percentage stays ≤ 100%.
+        if (inputTokens >= this.turnStats.inputTokens) {
+            this.turnStats.inputTokens = inputTokens;
+            this.turnStats.cachedTokens = cachedTokens;
+        }
+        this.turnStats.outputTokens += outputTokens;
+        this.turnStats.reasoningTokens += reasoningTokens;
         this.lastUsage = {
-            inputTokens: usage.input_tokens ?? 0,
-            outputTokens: usage.output_tokens ?? 0,
-            totalTokens: usage.total_tokens ?? 0,
-            cachedTokens: inputDetails.cached_tokens ?? 0,
-            reasoningTokens: outputDetails.reasoning_tokens ?? 0,
+            inputTokens: this.turnStats.inputTokens,
+            outputTokens: this.turnStats.outputTokens,
+            totalTokens: this.turnStats.inputTokens + this.turnStats.outputTokens,
+            cachedTokens: this.turnStats.cachedTokens,
+            reasoningTokens: this.turnStats.reasoningTokens,
             compactedTurns: this.compaction.compactedTurns,
         };
-        this.lastInputTokens = this.lastUsage.inputTokens;
+        // Keep lastInputTokens as the raw per-call value for compaction threshold checks.
+        this.lastInputTokens = inputTokens;
     }
     /**
      * Compact oversized tool outputs before storing them in rolling history.
@@ -485,6 +522,7 @@ export class MCPClientOpenAI {
         if (bypassSystemPrompt) {
             return this.processRawQuery(query, onThinking, abortSignal);
         }
+        this.startTurn();
         await this.ensureSystemPrompt();
         // Proactive compaction based on last real measured request.
         await this.maybeCompactHistory();
@@ -680,6 +718,7 @@ export class MCPClientOpenAI {
      * Raw mode: no cached instructions, no rolling history, no compaction state.
      */
     async processRawQuery(query, onThinking, abortSignal) {
+        this.startTurn();
         const tools = await this.buildTools();
         const isolatedHistory = [this.makeUserMessage(query)];
         let loopCount = 0;

package/dist/styles/MCPChat.css

@@ -442,6 +442,11 @@
   line-height: 1.6;
 }
 
+/* Reset white-space for markdown so ReactMarkdown HTML isn't double-spaced */
+.mcp-chat-message-content.markdown-content {
+  white-space: normal;
+}
+
 .mcp-chat-message-timestamp {
   font-size: 12px;
   opacity: 0.6;
@@ -470,7 +475,7 @@
 }
 
 /* Paragraphs */
-.markdown-content p { margin: 0 0 12px 0; }
+.markdown-content p { margin: 0 0 6px 0; }
 .markdown-content p:last-child { margin-bottom: 0; }
 
 /* Headings */
@@ -482,7 +487,7 @@
 .markdown-content h6 {
   font-weight: 700;
   line-height: 1.3;
-  margin: 16px 0 8px 0;
+  margin: 10px 0 2px 0;
   color: var(--mcp-text);
 }
 .markdown-content h1 { font-size: 1.4em; }
@@ -506,9 +511,12 @@
 
 /* Lists */
 .markdown-content ul,
-.markdown-content ol { margin: 0 0 12px 0; padding-left: 22px; }
-.markdown-content li { margin: 4px 0; }
-.markdown-content li > p { margin-bottom: 4px; }
+.markdown-content ol { margin: 0 0 6px 0; padding-left: 22px; }
+.markdown-content li { margin: 1px 0; }
+/* Loose lists: ReactMarkdown wraps li content in <p> tags when blank lines separate items.
+   Zero out all margins on these to prevent double-spacing. */
+.markdown-content li > p { margin: 0 !important; }
+.markdown-content li > p + p { margin-top: 2px !important; }
 /* Task lists (GFM) */
 .markdown-content input[type="checkbox"] {
   margin-right: 6px;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nqminds/mcp-client",
-  "version": "1.0.11",
+  "version": "1.0.14",
   "description": "Reusable MCP client component with AI chat interface",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",