@j0hanz/code-review-analyst-mcp 1.4.3 → 1.5.0

package/README.md

@@ -18,7 +18,7 @@ This server accepts unified diffs and returns structured JSON results — findin
 
 - **Impact Analysis** — Objective severity scoring, breaking change detection, and rollback complexity assessment.
 - **Review Summary** — Concise PR digest with merge recommendation and change statistics.
-- **Deep Code Inspection** — Pro model with 16K thinking budget for context-aware analysis using full file contents.
+- **Deep Code Inspection** — Pro model with high thinking level for context-aware analysis using full file contents.
 - **Search & Replace Fixes** — Verbatim, copy-paste-ready code fixes tied to specific findings.
 - **Test Plan Generation** — Systematic test case generation with priority ranking and pseudocode.
 - **Async Task Support** — All tools support MCP task lifecycle with progress notifications.
@@ -385,13 +385,13 @@ Create a test plan covering the changes in the diff using the Flash model with t
 
 ### Models
 
-| Tool                      | Model              | Thinking Budget |
-| ------------------------- | ------------------ | --------------- |
-| `analyze_pr_impact`       | `gemini-2.5-flash` | —               |
-| `generate_review_summary` | `gemini-2.5-flash` | —               |
-| `inspect_code_quality`    | `gemini-2.5-pro`   | 16,384 tokens   |
-| `suggest_search_replace`  | `gemini-2.5-pro`   | 16,384 tokens   |
-| `generate_test_plan`      | `gemini-2.5-flash` | 8,192 tokens    |
+| Tool                      | Model                    | Thinking Level |
+| ------------------------- | ------------------------ | -------------- |
+| `analyze_pr_impact`       | `gemini-3-flash-preview` | `minimal`      |
+| `generate_review_summary` | `gemini-3-flash-preview` | `minimal`      |
+| `inspect_code_quality`    | `gemini-3-pro-preview`   | `high`         |
+| `suggest_search_replace`  | `gemini-3-pro-preview`   | `high`         |
+| `generate_test_plan`      | `gemini-3-flash-preview` | `medium`       |
 
 ## Workflows
 

package/dist/lib/gemini.js

@@ -4,12 +4,12 @@ import { EventEmitter } from 'node:events';
 import { performance } from 'node:perf_hooks';
 import { setTimeout as sleep } from 'node:timers/promises';
 import { debuglog } from 'node:util';
-import { FinishReason, GoogleGenAI, HarmBlockThreshold, HarmCategory, } from '@google/genai';
+import { FinishReason, GoogleGenAI, HarmBlockThreshold, HarmCategory, ThinkingLevel, } from '@google/genai';
 import { createCachedEnvInt } from './env-config.js';
 import { getErrorMessage, RETRYABLE_UPSTREAM_ERROR_PATTERN } from './errors.js';
 // Lazy-cached: first call happens after parseCommandLineArgs() sets GEMINI_MODEL.
 let _defaultModel;
-const DEFAULT_MODEL = 'gemini-2.5-flash';
+const DEFAULT_MODEL = 'gemini-3-flash-preview';
 const GEMINI_MODEL_ENV_VAR = 'GEMINI_MODEL';
 const GEMINI_HARM_BLOCK_THRESHOLD_ENV_VAR = 'GEMINI_HARM_BLOCK_THRESHOLD';
 const GEMINI_INCLUDE_THOUGHTS_ENV_VAR = 'GEMINI_INCLUDE_THOUGHTS';
@@ -36,8 +36,8 @@ const DIGITS_ONLY_PATTERN = /^\d+$/;
 const SLEEP_UNREF_OPTIONS = { ref: false };
 const maxConcurrentCallsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS', 10);
 const concurrencyWaitMsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS_WAIT_MS', 2_000);
-const concurrencyPollMsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS_POLL_MS', 25);
 let activeCalls = 0;
+const slotWaiters = [];
 const RETRYABLE_TRANSIENT_CODES = new Set([
     'RESOURCE_EXHAUSTED',
     'UNAVAILABLE',
@@ -91,14 +91,31 @@ function parseSafetyThreshold(threshold) {
     }
     return SAFETY_THRESHOLD_BY_NAME[normalizedThreshold];
 }
-function getThinkingConfig(thinkingBudget, includeThoughts) {
-    if (thinkingBudget === undefined) {
+function getThinkingConfig(thinkingLevel, includeThoughts) {
+    if (thinkingLevel === undefined && !includeThoughts) {
         return undefined;
     }
+    const config = {};
+    if (thinkingLevel !== undefined) {
+        switch (thinkingLevel) {
+            case 'minimal':
+                config.thinkingLevel = ThinkingLevel.MINIMAL;
+                break;
+            case 'low':
+                config.thinkingLevel = ThinkingLevel.LOW;
+                break;
+            case 'medium':
+                config.thinkingLevel = ThinkingLevel.MEDIUM;
+                break;
+            case 'high':
+                config.thinkingLevel = ThinkingLevel.HIGH;
+                break;
+        }
+    }
     if (includeThoughts) {
-        return { includeThoughts: true, thinkingBudget };
+        config.includeThoughts = true;
     }
-    return { thinkingBudget };
+    return config;
 }
 function parseBooleanEnv(value) {
     const normalized = value.trim().toLowerCase();
@@ -300,9 +317,9 @@ function getRetryDelayMs(attempt) {
 }
 function buildGenerationConfig(request, abortSignal) {
     const includeThoughts = request.includeThoughts ?? getDefaultIncludeThoughts();
-    const thinkingConfig = getThinkingConfig(request.thinkingBudget, includeThoughts);
+    const thinkingConfig = getThinkingConfig(request.thinkingLevel, includeThoughts);
     const config = {
-        temperature: request.temperature ?? 0.2,
+        temperature: request.temperature ?? 1.0,
         maxOutputTokens: request.maxOutputTokens ?? DEFAULT_MAX_OUTPUT_TOKENS,
         responseMimeType: 'application/json',
         responseSchema: request.responseSchema,
@@ -443,19 +460,62 @@ async function runWithRetries(request, model, timeoutMs, maxRetries, onLog) {
 function canRetryAttempt(attempt, maxRetries, error) {
     return attempt < maxRetries && shouldRetry(error);
 }
+function tryWakeNextWaiter() {
+    const next = slotWaiters.shift();
+    if (next !== undefined) {
+        next();
+    }
+}
 async function waitForConcurrencySlot(limit, requestSignal) {
+    if (activeCalls < limit) {
+        return;
+    }
+    if (requestSignal?.aborted) {
+        throw new Error('Gemini request was cancelled.');
+    }
     const waitLimitMs = concurrencyWaitMsConfig.get();
-    const pollMs = concurrencyPollMsConfig.get();
-    const deadline = performance.now() + waitLimitMs;
-    while (activeCalls >= limit) {
-        if (requestSignal?.aborted) {
-            throw new Error('Gemini request was cancelled.');
-        }
-        if (performance.now() >= deadline) {
-            throw new Error(formatConcurrencyLimitErrorMessage(limit, waitLimitMs));
+    return new Promise((resolve, reject) => {
+        let settled = false;
+        const waiter = () => {
+            if (settled)
+                return;
+            settled = true;
+            clearTimeout(deadlineTimer);
+            if (requestSignal) {
+                requestSignal.removeEventListener('abort', onAbort);
+            }
+            resolve();
+        };
+        slotWaiters.push(waiter);
+        const deadlineTimer = setTimeout(() => {
+            if (settled)
+                return;
+            settled = true;
+            const idx = slotWaiters.indexOf(waiter);
+            if (idx !== -1) {
+                slotWaiters.splice(idx, 1);
+            }
+            if (requestSignal) {
+                requestSignal.removeEventListener('abort', onAbort);
+            }
+            reject(new Error(formatConcurrencyLimitErrorMessage(limit, waitLimitMs)));
+        }, waitLimitMs);
+        deadlineTimer.unref();
+        const onAbort = () => {
+            if (settled)
+                return;
+            settled = true;
+            const idx = slotWaiters.indexOf(waiter);
+            if (idx !== -1) {
+                slotWaiters.splice(idx, 1);
+            }
+            clearTimeout(deadlineTimer);
+            reject(new Error('Gemini request was cancelled.'));
+        };
+        if (requestSignal) {
+            requestSignal.addEventListener('abort', onAbort, { once: true });
         }
-        await sleep(pollMs, undefined, SLEEP_UNREF_OPTIONS);
-    }
+    });
 }
 export async function generateStructuredJson(request) {
     const model = request.model ?? getDefaultModel();
@@ -470,5 +530,6 @@ export async function generateStructuredJson(request) {
     }
     finally {
         activeCalls -= 1;
+        tryWakeNextWaiter();
     }
 }

package/dist/lib/model-config.d.ts

@@ -1,29 +1,39 @@
 /** Fast, cost-effective model for summarization and light analysis. */
-export declare const FLASH_MODEL = "gemini-2.5-flash";
+export declare const FLASH_MODEL = "gemini-3-flash-preview";
 /** High-capability model for deep reasoning, quality inspection, and reliable code generation. */
-export declare const PRO_MODEL = "gemini-2.5-pro";
-/** Thinking budget (tokens) for Flash model thinking tasks (test plans, search/replace). */
-export declare const FLASH_THINKING_BUDGET: 8192;
-/** Thinking budget (tokens) for Pro model deep-analysis tasks (code quality inspection). */
-export declare const PRO_THINKING_BUDGET: 16384;
-/** Output cap for Flash triage tools (impact, summary). */
-export declare const FLASH_TRIAGE_MAX_OUTPUT_TOKENS: 4096;
-/** Output cap for API breaking-change detection (migration guidance needs room). */
-export declare const FLASH_API_BREAKING_MAX_OUTPUT_TOKENS: 4096;
-/** Output cap for test-plan generation (includes pseudocode snippets). */
-export declare const FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS: 4096;
-/** Output cap for Pro deep review findings. */
-export declare const PRO_REVIEW_MAX_OUTPUT_TOKENS: 8192;
-/** Output cap for Pro search/replace remediation blocks. */
-export declare const PRO_PATCH_MAX_OUTPUT_TOKENS: 4096;
-/** Output cap for Flash complexity analysis reports. */
-export declare const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS: 2048;
-/** Extended timeout for Pro model calls (ms). Pro thinks longer than Flash. */
+export declare const PRO_MODEL = "gemini-3-pro-preview";
+/** Default language hint. */
+export declare const DEFAULT_LANGUAGE = "detect";
+/** Default test-framework hint. */
+export declare const DEFAULT_FRAMEWORK = "detect";
+/** Extended timeout for Pro model calls (ms). */
 export declare const DEFAULT_TIMEOUT_PRO_MS = 120000;
 export declare const MODEL_TIMEOUT_MS: {
     readonly defaultPro: 120000;
 };
-/** Default language hint when not specified by the user. Tells the model to auto-detect. */
-export declare const DEFAULT_LANGUAGE = "detect";
-/** Default test-framework hint when not specified by the user. Tells the model to auto-detect. */
-export declare const DEFAULT_FRAMEWORK = "detect";
+/** Thinking level for Flash triage. */
+export declare const FLASH_TRIAGE_THINKING_LEVEL: "minimal";
+/** Thinking level for Flash analysis. */
+export declare const FLASH_THINKING_LEVEL: "medium";
+/** Thinking level for Pro deep analysis. */
+export declare const PRO_THINKING_LEVEL: "high";
+/** Output cap for Flash API breaking-change detection. */
+export declare const FLASH_API_BREAKING_MAX_OUTPUT_TOKENS: 4096;
+/** Output cap for Flash complexity analysis. */
+export declare const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS: 2048;
+/** Output cap for Flash test-plan generation. */
+export declare const FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS: 8192;
+/** Output cap for Flash triage tools. */
+export declare const FLASH_TRIAGE_MAX_OUTPUT_TOKENS: 4096;
+/** Output cap for Pro patch generation. */
+export declare const PRO_PATCH_MAX_OUTPUT_TOKENS: 8192;
+/** Output cap for Pro deep review findings. */
+export declare const PRO_REVIEW_MAX_OUTPUT_TOKENS: 12288;
+/** Temperature for analytical tools. */
+export declare const ANALYSIS_TEMPERATURE: 1;
+/** Temperature for creative synthesis (test plans). */
+export declare const CREATIVE_TEMPERATURE: 1;
+/** Temperature for code patch generation. */
+export declare const PATCH_TEMPERATURE: 1;
+/** Temperature for triage/classification tools. */
+export declare const TRIAGE_TEMPERATURE: 1;

package/dist/lib/model-config.js

@@ -1,43 +1,71 @@
 /** Fast, cost-effective model for summarization and light analysis. */
-export const FLASH_MODEL = 'gemini-2.5-flash';
+export const FLASH_MODEL = 'gemini-3-flash-preview';
 /** High-capability model for deep reasoning, quality inspection, and reliable code generation. */
-export const PRO_MODEL = 'gemini-2.5-pro';
-const THINKING_BUDGET_TOKENS = {
-    flash: 8_192,
-    pro: 16_384,
+export const PRO_MODEL = 'gemini-3-pro-preview';
+/** Default hint for auto-detection. */
+const DEFAULT_DETECT_HINT = 'detect';
+/** Default language hint. */
+export const DEFAULT_LANGUAGE = DEFAULT_DETECT_HINT;
+/** Default test-framework hint. */
+export const DEFAULT_FRAMEWORK = DEFAULT_DETECT_HINT;
+/** Extended timeout for Pro model calls (ms). */
+export const DEFAULT_TIMEOUT_PRO_MS = 120_000;
+export const MODEL_TIMEOUT_MS = {
+    defaultPro: DEFAULT_TIMEOUT_PRO_MS,
 };
+Object.freeze(MODEL_TIMEOUT_MS);
+// ---------------------------------------------------------------------------
+// Budgets (Thinking & Output)
+// ---------------------------------------------------------------------------
+const THINKING_LEVELS = {
+    /** Minimal thinking for triage/classification. */
+    flashTriage: 'minimal',
+    /** Medium thinking for analysis tasks. */
+    flash: 'medium',
+    /** High thinking for deep review and patches. */
+    pro: 'high',
+};
+// Thinking budget in tokens for Flash and Pro tools. Note that these are not hard limits, but rather guidelines to encourage concise responses and manage latency/cost.
 const OUTPUT_TOKEN_BUDGET = {
-    flashTriage: 4_096,
-    flashTestPlan: 4_096,
     flashApiBreaking: 4_096,
     flashComplexity: 2_048,
-    proReview: 8_192,
-    proPatch: 4_096,
+    flashTestPlan: 8_192,
+    flashTriage: 4_096,
+    proPatch: 8_192,
+    proReview: 12_288,
 };
-const DEFAULT_DETECT_HINT = 'detect';
-/** Thinking budget (tokens) for Flash model thinking tasks (test plans, search/replace). */
-export const FLASH_THINKING_BUDGET = THINKING_BUDGET_TOKENS.flash;
-/** Thinking budget (tokens) for Pro model deep-analysis tasks (code quality inspection). */
-export const PRO_THINKING_BUDGET = THINKING_BUDGET_TOKENS.pro;
-/** Output cap for Flash triage tools (impact, summary). */
-export const FLASH_TRIAGE_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashTriage;
-/** Output cap for API breaking-change detection (migration guidance needs room). */
+/** Thinking level for Flash triage. */
+export const FLASH_TRIAGE_THINKING_LEVEL = THINKING_LEVELS.flashTriage;
+/** Thinking level for Flash analysis. */
+export const FLASH_THINKING_LEVEL = THINKING_LEVELS.flash;
+/** Thinking level for Pro deep analysis. */
+export const PRO_THINKING_LEVEL = THINKING_LEVELS.pro;
+/** Output cap for Flash API breaking-change detection. */
 export const FLASH_API_BREAKING_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashApiBreaking;
-/** Output cap for test-plan generation (includes pseudocode snippets). */
+/** Output cap for Flash complexity analysis. */
+export const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashComplexity;
+/** Output cap for Flash test-plan generation. */
 export const FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashTestPlan;
+/** Output cap for Flash triage tools. */
+export const FLASH_TRIAGE_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashTriage;
+/** Output cap for Pro patch generation. */
+export const PRO_PATCH_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.proPatch;
 /** Output cap for Pro deep review findings. */
 export const PRO_REVIEW_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.proReview;
-/** Output cap for Pro search/replace remediation blocks. */
-export const PRO_PATCH_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.proPatch;
-/** Output cap for Flash complexity analysis reports. */
-export const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashComplexity;
-/** Extended timeout for Pro model calls (ms). Pro thinks longer than Flash. */
-export const DEFAULT_TIMEOUT_PRO_MS = 120_000;
-export const MODEL_TIMEOUT_MS = {
-    defaultPro: DEFAULT_TIMEOUT_PRO_MS,
+// ---------------------------------------------------------------------------
+// Temperatures
+// ---------------------------------------------------------------------------
+const TOOL_TEMPERATURE = {
+    analysis: 1.0, // Gemini 3 recommends 1.0 for all tasks
+    creative: 1.0, // Gemini 3 recommends 1.0 for all tasks
+    patch: 1.0, // Gemini 3 recommends 1.0 for all tasks
+    triage: 1.0, // Gemini 3 recommends 1.0 for all tasks
 };
-Object.freeze(MODEL_TIMEOUT_MS);
-/** Default language hint when not specified by the user. Tells the model to auto-detect. */
-export const DEFAULT_LANGUAGE = DEFAULT_DETECT_HINT;
-/** Default test-framework hint when not specified by the user. Tells the model to auto-detect. */
-export const DEFAULT_FRAMEWORK = DEFAULT_DETECT_HINT;
+/** Temperature for analytical tools. */
+export const ANALYSIS_TEMPERATURE = TOOL_TEMPERATURE.analysis;
+/** Temperature for creative synthesis (test plans). */
+export const CREATIVE_TEMPERATURE = TOOL_TEMPERATURE.creative;
+/** Temperature for code patch generation. */
+export const PATCH_TEMPERATURE = TOOL_TEMPERATURE.patch;
+/** Temperature for triage/classification tools. */
+export const TRIAGE_TEMPERATURE = TOOL_TEMPERATURE.triage;

package/dist/lib/tool-contracts.d.ts

@@ -13,9 +13,16 @@ export interface ToolContract {
     model: string;
     /** Set to 0 for synchronous (non-Gemini) tools. */
     timeoutMs: number;
-    thinkingBudget?: number;
+    thinkingLevel?: 'minimal' | 'low' | 'medium' | 'high';
     /** Set to 0 for synchronous (non-Gemini) tools. */
     maxOutputTokens: number;
+    /**
+     * Sampling temperature for the Gemini call.
+     * Lower values (0.0–0.1) favour deterministic structured output;
+     * higher values (0.2) add diversity for creative synthesis tasks.
+     * Omit to use the global default (0.2).
+     */
+    temperature?: number;
     params: readonly ToolParameterContract[];
     outputShape: string;
     gotchas: readonly string[];
@@ -41,9 +48,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "analyze_pr_impact";
     readonly purpose: "Assess severity, categories, breaking changes, and rollback complexity.";
-    readonly model: "gemini-2.5-flash";
+    readonly model: "gemini-3-flash-preview";
     readonly timeoutMs: 90000;
+    readonly thinkingLevel: "minimal";
     readonly maxOutputTokens: 4096;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -63,9 +72,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "generate_review_summary";
     readonly purpose: "Produce PR summary, risk rating, and merge recommendation.";
-    readonly model: "gemini-2.5-flash";
+    readonly model: "gemini-3-flash-preview";
     readonly timeoutMs: 90000;
+    readonly thinkingLevel: "minimal";
     readonly maxOutputTokens: 4096;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -85,10 +96,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "inspect_code_quality";
     readonly purpose: "Deep code review with optional full-file context.";
-    readonly model: "gemini-2.5-pro";
+    readonly model: "gemini-3-pro-preview";
     readonly timeoutMs: 120000;
-    readonly thinkingBudget: 16384;
-    readonly maxOutputTokens: 8192;
+    readonly thinkingLevel: "high";
+    readonly maxOutputTokens: 12288;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -127,10 +139,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "suggest_search_replace";
     readonly purpose: "Generate verbatim search/replace fix blocks for one finding.";
-    readonly model: "gemini-2.5-pro";
+    readonly model: "gemini-3-pro-preview";
     readonly timeoutMs: 120000;
-    readonly thinkingBudget: 16384;
-    readonly maxOutputTokens: 4096;
+    readonly thinkingLevel: "high";
+    readonly maxOutputTokens: 8192;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "findingTitle";
         readonly type: "string";
@@ -151,10 +164,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "generate_test_plan";
     readonly purpose: "Generate prioritized test cases and coverage guidance.";
-    readonly model: "gemini-2.5-flash";
+    readonly model: "gemini-3-flash-preview";
     readonly timeoutMs: 90000;
-    readonly thinkingBudget: 8192;
-    readonly maxOutputTokens: 4096;
+    readonly thinkingLevel: "medium";
+    readonly maxOutputTokens: 8192;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -186,10 +200,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "analyze_time_space_complexity";
     readonly purpose: "Analyze Big-O complexity and detect degradations in changed code.";
-    readonly model: "gemini-2.5-flash";
+    readonly model: "gemini-3-flash-preview";
     readonly timeoutMs: 90000;
-    readonly thinkingBudget: 8192;
+    readonly thinkingLevel: "medium";
     readonly maxOutputTokens: 2048;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "language";
         readonly type: "string";
@@ -203,9 +218,11 @@ export declare const TOOL_CONTRACTS: readonly [{
 }, {
     readonly name: "detect_api_breaking_changes";
     readonly purpose: "Detect breaking API/interface changes in a diff.";
-    readonly model: "gemini-2.5-flash";
+    readonly model: "gemini-3-flash-preview";
     readonly timeoutMs: 90000;
+    readonly thinkingLevel: "minimal";
     readonly maxOutputTokens: 4096;
+    readonly temperature: 1;
     readonly params: readonly [{
         readonly name: "language";
         readonly type: "string";

package/dist/lib/tool-contracts.js

@@ -1,4 +1,4 @@
-import { DEFAULT_TIMEOUT_PRO_MS, FLASH_API_BREAKING_MAX_OUTPUT_TOKENS, FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS, FLASH_MODEL, FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS, FLASH_THINKING_BUDGET, FLASH_TRIAGE_MAX_OUTPUT_TOKENS, PRO_MODEL, PRO_PATCH_MAX_OUTPUT_TOKENS, PRO_REVIEW_MAX_OUTPUT_TOKENS, PRO_THINKING_BUDGET, } from './model-config.js';
+import { ANALYSIS_TEMPERATURE, CREATIVE_TEMPERATURE, DEFAULT_TIMEOUT_PRO_MS, FLASH_API_BREAKING_MAX_OUTPUT_TOKENS, FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS, FLASH_MODEL, FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS, FLASH_THINKING_LEVEL, FLASH_TRIAGE_MAX_OUTPUT_TOKENS, FLASH_TRIAGE_THINKING_LEVEL, PATCH_TEMPERATURE, PRO_MODEL, PRO_PATCH_MAX_OUTPUT_TOKENS, PRO_REVIEW_MAX_OUTPUT_TOKENS, PRO_THINKING_LEVEL, TRIAGE_TEMPERATURE, } from './model-config.js';
 const DEFAULT_TIMEOUT_FLASH_MS = 90_000;
 export const INSPECTION_FOCUS_AREAS = [
     'security',
@@ -40,7 +40,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Assess severity, categories, breaking changes, and rollback complexity.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingLevel: FLASH_TRIAGE_THINKING_LEVEL,
         maxOutputTokens: FLASH_TRIAGE_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -71,7 +73,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Produce PR summary, risk rating, and merge recommendation.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingLevel: FLASH_TRIAGE_THINKING_LEVEL,
         maxOutputTokens: FLASH_TRIAGE_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -102,8 +106,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Deep code review with optional full-file context.',
         model: PRO_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_PRO_MS,
-        thinkingBudget: PRO_THINKING_BUDGET,
+        thinkingLevel: PRO_THINKING_LEVEL,
         maxOutputTokens: PRO_REVIEW_MAX_OUTPUT_TOKENS,
+        temperature: ANALYSIS_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -158,8 +163,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Generate verbatim search/replace fix blocks for one finding.',
         model: PRO_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_PRO_MS,
-        thinkingBudget: PRO_THINKING_BUDGET,
+        thinkingLevel: PRO_THINKING_LEVEL,
         maxOutputTokens: PRO_PATCH_MAX_OUTPUT_TOKENS,
+        temperature: PATCH_TEMPERATURE,
         params: [
             {
                 name: 'findingTitle',
@@ -192,8 +198,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Generate prioritized test cases and coverage guidance.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
-        thinkingBudget: FLASH_THINKING_BUDGET,
+        thinkingLevel: FLASH_THINKING_LEVEL,
         maxOutputTokens: FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS,
+        temperature: CREATIVE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -238,8 +245,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Analyze Big-O complexity and detect degradations in changed code.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
-        thinkingBudget: FLASH_THINKING_BUDGET,
+        thinkingLevel: FLASH_THINKING_LEVEL,
         maxOutputTokens: FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS,
+        temperature: ANALYSIS_TEMPERATURE,
         params: [
             {
                 name: 'language',
@@ -261,7 +269,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Detect breaking API/interface changes in a diff.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingLevel: FLASH_TRIAGE_THINKING_LEVEL,
         maxOutputTokens: FLASH_API_BREAKING_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'language',

package/dist/lib/tool-factory.d.ts

@@ -55,14 +55,21 @@ export interface StructuredToolTaskConfig<TInput extends object = Record<string,
     transformResult?: (input: TInput, result: TResult, ctx: ToolExecutionContext) => TFinal;
     /** Optional validation hook for input parameters. */
     validateInput?: (input: TInput, ctx: ToolExecutionContext) => Promise<ReturnType<typeof createErrorToolResponse> | undefined> | ReturnType<typeof createErrorToolResponse> | undefined;
-    /** Optional Gemini model to use (e.g. 'gemini-2.5-pro'). */
+    /** Optional Gemini model to use (e.g. 'gemini-3-pro-preview'). */
     model?: string;
-    /** Optional thinking budget in tokens. */
-    thinkingBudget?: number;
+    /** Optional thinking level. */
+    thinkingLevel?: 'minimal' | 'low' | 'medium' | 'high';
     /** Optional timeout in ms for the Gemini call. Defaults to 90,000 ms. Use DEFAULT_TIMEOUT_PRO_MS for Pro model calls. */
     timeoutMs?: number;
     /** Optional max output tokens for Gemini. */
     maxOutputTokens?: number;
+    /**
+     * Optional sampling temperature for this tool's Gemini call.
+     * Lower values (0.0–0.1) favour determinism for structured extraction;
+     * higher values (0.2) add useful diversity for creative synthesis tasks.
+     * Falls back to the global default (0.2) when omitted.
+     */
+    temperature?: number;
     /** Optional opt-in to Gemini thought output. Defaults to false. */
     includeThoughts?: boolean;
     /** Optional formatter for human-readable text output. */

package/dist/lib/tool-factory.js

@@ -31,8 +31,8 @@ function createGenerationRequest(config, promptParts, responseSchema, onLog, sig
     if (config.model !== undefined) {
         request.model = config.model;
     }
-    if (config.thinkingBudget !== undefined) {
-        request.thinkingBudget = config.thinkingBudget;
+    if (config.thinkingLevel !== undefined) {
+        request.thinkingLevel = config.thinkingLevel;
     }
     if (config.timeoutMs !== undefined) {
         request.timeoutMs = config.timeoutMs;
@@ -40,6 +40,9 @@ function createGenerationRequest(config, promptParts, responseSchema, onLog, sig
     if (config.maxOutputTokens !== undefined) {
         request.maxOutputTokens = config.maxOutputTokens;
     }
+    if (config.temperature !== undefined) {
+        request.temperature = config.temperature;
+    }
     if (config.includeThoughts !== undefined) {
         request.includeThoughts = config.includeThoughts;
     }

package/dist/lib/types.d.ts

@@ -5,7 +5,7 @@ export interface GeminiRequestExecutionOptions {
     timeoutMs?: number;
     temperature?: number;
     maxOutputTokens?: number;
-    thinkingBudget?: number;
+    thinkingLevel?: 'minimal' | 'low' | 'medium' | 'high';
     includeThoughts?: boolean;
     signal?: AbortSignal;
     onLog?: GeminiLogHandler;

package/dist/prompts/index.js

@@ -43,9 +43,9 @@ function getToolGuide(tool) {
     if (!contract) {
         return `Use \`${tool}\` to analyze your code changes.`;
     }
-    const { thinkingBudget } = contract;
-    const modelLine = thinkingBudget !== undefined
-        ? `Model: ${contract.model} (thinking budget ${thinkingBudget}, output cap ${contract.maxOutputTokens}).`
+    const { thinkingLevel } = contract;
+    const modelLine = thinkingLevel !== undefined
+        ? `Model: ${contract.model} (thinking level ${thinkingLevel}, output cap ${contract.maxOutputTokens}).`
         : `Model: ${contract.model} (output cap ${contract.maxOutputTokens}).`;
     return `Tool: ${contract.name}\n${modelLine}\nOutput: ${contract.outputShape}\nUse: ${contract.purpose}`;
 }

package/dist/resources/instructions.js

@@ -23,9 +23,9 @@ function formatToolSection(contract) {
 ${parameterLines.join('\n')}
 - Output shape: \`${contract.outputShape}\``;
     }
-    const thinkingLine = contract.thinkingBudget === undefined
-        ? '- Thinking budget: disabled'
-        : `- Thinking budget: ${contract.thinkingBudget}`;
+    const thinkingLine = contract.thinkingLevel === undefined
+        ? '- Thinking level: disabled'
+        : `- Thinking level: ${contract.thinkingLevel}`;
     return `### \`${contract.name}\`
 - Purpose: ${contract.purpose}
 - Model: \`${contract.model}\`

package/dist/resources/server-config.js

@@ -24,8 +24,8 @@ function formatNumber(value) {
 function formatTimeout(ms) {
     return `${Math.round(ms / 1_000)}s`;
 }
-function formatThinkingBudget(budget) {
-    return budget !== undefined ? formatNumber(budget) : '—';
+function formatThinkingLevel(level) {
+    return level ?? '—';
 }
 export function buildServerConfig() {
     const maxDiffChars = diffCharsConfig.get();
@@ -37,7 +37,7 @@ export function buildServerConfig() {
     const toolRows = getToolContracts()
         .filter((contract) => contract.model !== 'none')
         .map((contract) => {
-        return `| \`${contract.name}\` | \`${contract.model}\` | ${formatThinkingBudget(contract.thinkingBudget)} | ${formatTimeout(contract.timeoutMs)} | ${formatNumber(contract.maxOutputTokens)} |`;
+        return `| \`${contract.name}\` | \`${contract.model}\` | ${formatThinkingLevel(contract.thinkingLevel)} | ${formatTimeout(contract.timeoutMs)} | ${formatNumber(contract.maxOutputTokens)} |`;
     })
         .join('\n');
     return `# Server Configuration
@@ -55,7 +55,7 @@ export function buildServerConfig() {
 
 Default model: \`${defaultModel}\` (override with \`GEMINI_MODEL\`)
 
-| Tool | Model | Thinking Budget | Timeout | Max Output Tokens |
+| Tool | Model | Thinking Level | Timeout | Max Output Tokens |
 |------|-------|----------------|---------|-------------------|
 ${toolRows}
 

package/dist/resources/tool-info.js

@@ -10,8 +10,8 @@ function formatNumber(value) {
 function formatTimeout(timeoutMs) {
     return `${Math.round(timeoutMs / 1_000)}s`;
 }
-function formatThinkingBudget(thinkingBudget) {
-    return thinkingBudget === undefined ? '-' : formatNumber(thinkingBudget);
+function formatThinkingLevel(thinkingLevel) {
+    return thinkingLevel ?? '-';
 }
 function formatOutputTokens(maxOutputTokens) {
     return formatNumber(maxOutputTokens);
@@ -29,7 +29,7 @@ function toToolInfoEntry(contract) {
         name: contract.name,
         purpose: contract.purpose,
         model: contract.model,
-        thinkingBudget: formatThinkingBudget(contract.thinkingBudget),
+        thinkingLevel: formatThinkingLevel(contract.thinkingLevel),
         timeout: formatTimeout(contract.timeoutMs),
         maxOutputTokens: formatOutputTokens(contract.maxOutputTokens),
         params: parameterRows.join('\n'),
@@ -62,7 +62,7 @@ function formatToolInfo(entry) {
 ${entry.purpose}
 
 ## Model
-\`${entry.model}\` (thinking budget: ${entry.thinkingBudget}, timeout: ${entry.timeout}, max output tokens: ${entry.maxOutputTokens})
+\`${entry.model}\` (thinking level: ${entry.thinkingLevel}, timeout: ${entry.timeout}, max output tokens: ${entry.maxOutputTokens})
 
 ## Parameters
 ${entry.params}

package/dist/schemas/outputs.d.ts

@@ -18,9 +18,9 @@ export declare const DefaultOutputSchema: z.ZodObject<{
 }, z.core.$strict>;
 export declare const ReviewFindingSchema: z.ZodObject<{
     severity: z.ZodEnum<{
-        low: "low";
         medium: "medium";
         high: "high";
+        low: "low";
         critical: "critical";
     }>;
     file: z.ZodString;
@@ -31,9 +31,9 @@ export declare const ReviewFindingSchema: z.ZodObject<{
 }, z.core.$strict>;
 export declare const PrImpactResultSchema: z.ZodObject<{
     severity: z.ZodEnum<{
-        low: "low";
         medium: "medium";
         high: "high";
+        low: "low";
         critical: "critical";
     }>;
     categories: z.ZodArray<z.ZodEnum<{
@@ -61,9 +61,9 @@ export declare const PrImpactResultSchema: z.ZodObject<{
 export declare const ReviewSummaryResultSchema: z.ZodObject<{
     summary: z.ZodString;
     overallRisk: z.ZodEnum<{
-        low: "low";
         medium: "medium";
         high: "high";
+        low: "low";
     }>;
     keyChanges: z.ZodArray<z.ZodString>;
     recommendation: z.ZodString;
@@ -76,16 +76,16 @@ export declare const ReviewSummaryResultSchema: z.ZodObject<{
 export declare const CodeQualityResultSchema: z.ZodObject<{
     summary: z.ZodString;
     overallRisk: z.ZodEnum<{
-        low: "low";
         medium: "medium";
         high: "high";
+        low: "low";
         critical: "critical";
     }>;
     findings: z.ZodArray<z.ZodObject<{
         severity: z.ZodEnum<{
-            low: "low";
             medium: "medium";
             high: "high";
+            low: "low";
             critical: "critical";
         }>;
         file: z.ZodString;
@@ -101,16 +101,16 @@ export declare const CodeQualityOutputSchema: z.ZodObject<{
     totalFindings: z.ZodOptional<z.ZodNumber>;
     summary: z.ZodString;
     overallRisk: z.ZodEnum<{
-        low: "low";
         medium: "medium";
         high: "high";
+        low: "low";
         critical: "critical";
     }>;
     findings: z.ZodArray<z.ZodObject<{
         severity: z.ZodEnum<{
-            low: "low";
             medium: "medium";
             high: "high";
+            low: "low";
             critical: "critical";
         }>;
         file: z.ZodString;

package/dist/tools/analyze-complexity.js

@@ -23,8 +23,11 @@ export function registerAnalyzeComplexityTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
-        ...(TOOL_CONTRACT.thinkingBudget !== undefined
-            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
             : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
@@ -43,7 +46,7 @@ export function registerAnalyzeComplexityTool(server) {
                 : '';
             return {
                 systemInstruction: SYSTEM_INSTRUCTION,
-                prompt: `${languageLine}\nDiff:\n${diff}`.trimStart(),
+                prompt: `${languageLine}\\nDiff:\\n${diff}\\n\\nBased on the diff above, analyze the Big-O time and space complexity.`.trimStart(),
             };
         },
     });

package/dist/tools/analyze-pr-impact.js

@@ -27,6 +27,12 @@ export function registerAnalyzePrImpactTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)
@@ -50,6 +56,8 @@ ${fileSummary}
 
 Diff:
 ${diff}
+
+Based on the diff and change stats above, analyze the PR impact.
 `,
             };
         },

package/dist/tools/detect-api-breaking.js

@@ -23,6 +23,12 @@ export function registerDetectApiBreakingTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)
@@ -40,7 +46,7 @@ export function registerDetectApiBreakingTool(server) {
                 : '';
             return {
                 systemInstruction: SYSTEM_INSTRUCTION,
-                prompt: `${languageLine}\nDiff:\n${diff}`.trimStart(),
+                prompt: `${languageLine}\\nDiff:\\n${diff}\\n\\nBased on the diff above, detect any breaking API changes.`.trimStart(),
             };
         },
     });

package/dist/tools/generate-review-summary.js

@@ -34,6 +34,12 @@ export function registerGenerateReviewSummaryTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)
@@ -64,6 +70,8 @@ Stats: ${files} files, +${added}, -${deleted}
 
 Diff:
 ${diff}
+
+Based on the diff and stats above, summarize the PR and provide a merge recommendation.
 `,
             };
         },

package/dist/tools/generate-test-plan.js

@@ -27,8 +27,11 @@ export function registerGenerateTestPlanTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
-        ...(TOOL_CONTRACT.thinkingBudget !== undefined
-            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
             : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
@@ -58,6 +61,8 @@ Changed Files: ${paths.join(', ')}
 
 Diff:
 ${diff}
+
+Based on the diff and stats above, generate an actionable test plan.
 `,
             };
         },

package/dist/tools/inspect-code-quality.js

@@ -63,8 +63,11 @@ export function registerInspectCodeQualityTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
-        ...(TOOL_CONTRACT.thinkingBudget !== undefined
-            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
             : undefined),
         progressContext: (input) => {
             const fileCount = input.files?.length;
@@ -114,6 +117,8 @@ ${fileSummary}
 Diff:
 ${diff}
 ${fileContext}
+
+Based on the diff and file context above, perform a deep code review focusing on the specified areas.
 `,
             };
         },

package/dist/tools/suggest-search-replace.js

@@ -27,8 +27,11 @@ export function registerSuggestSearchReplaceTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
-        ...(TOOL_CONTRACT.thinkingBudget !== undefined
-            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+        ...(TOOL_CONTRACT.thinkingLevel !== undefined
+            ? { thinkingLevel: TOOL_CONTRACT.thinkingLevel }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
             : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
@@ -55,6 +58,8 @@ Changed Files: ${paths.join(', ')}
 
 Diff:
 ${diff}
+
+Based on the diff and finding details above, generate minimal search-and-replace blocks to fix the issue.
 `,
             };
         },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@j0hanz/code-review-analyst-mcp",
-  "version": "1.4.3",
+  "version": "1.5.0",
   "mcpName": "io.github.j0hanz/code-review-analyst",
   "description": "Gemini-powered MCP server for code review analysis.",
   "type": "module",