converse-mcp-server 2.22.8 → 2.26.1

package/docs/API.md

@@ -2,12 +2,13 @@
 
 ## Overview
 
-The Converse MCP Server provides four main tools through the Model Context Protocol (MCP):
+The Converse MCP Server provides five main tools through the Model Context Protocol (MCP):
 
 1. **Chat Tool** - Single-provider conversational AI with context support and AI summarization
 2. **Consensus Tool** - Multi-provider parallel execution with response aggregation and combined summaries
-3. **Check Status Tool** - Monitor and retrieve results from asynchronous operations with intelligent summaries
-4. **Cancel Job Tool** - Cancel running background operations
+3. **Conversation Tool** - Turn-based multi-model round-table where models respond sequentially, each seeing the full running transcript
+4. **Check Status Tool** - Monitor and retrieve results from asynchronous operations with intelligent summaries
+5. **Cancel Job Tool** - Cancel running background operations
 
 All tools support both **synchronous** (immediate response) and **asynchronous** (background processing) execution modes. When AI summarization is enabled, tools automatically generate titles and summaries for better context understanding.
 
@@ -350,6 +351,165 @@ conv_architecture_design/
 }
 ```
 
+### Conversation Tool
+
+**Description**: Turn-based multi-model round-table. Unlike consensus (parallel, all models answer the same prompt), models here respond **sequentially in the order given**, and each model sees the full running transcript of every turn before it. One tool call runs exactly **one lap** (one turn per model). The caller drives more laps by passing back the returned `continuation_id`; every lap appends to one shared, accumulating transcript that all models see.
+
+#### Request Schema
+
+```json
+{
+  "type": "object",
+  "properties": {
+    "prompt": {
+      "type": "string",
+      "description": "The topic or question to open the round-table with. Example: 'Critique this caching strategy and propose improvements.'"
+    },
+    "models": {
+      "type": "array",
+      "items": {"type": "string"},
+      "minItems": 1,
+      "description": "Ordered list of models. ORDER MATTERS: models speak one after another in this exact order, each seeing the transcript of those before it. Example: ['codex', 'gemini', 'claude']"
+    },
+    "continuation_id": {
+      "type": "string",
+      "description": "Thread continuation ID for running more laps. Auto-generated in the first response; pass it back to run another lap where every model again sees the full accumulated transcript. You MAY change the models list on a resuming lap."
+    },
+    "turn_prompt": {
+      "type": "string",
+      "description": "Optional custom per-turn instruction appended to the round-table framing each model receives. Example: 'Focus on security implications in your turn.'"
+    },
+    "files": {
+      "type": "array",
+      "items": {"type": "string"},
+      "description": "File paths shared with every participant in the lap. Supports line ranges: file.txt{10:50}."
+    },
+    "images": {
+      "type": "array",
+      "items": {"type": "string"},
+      "description": "Image paths for visual context (absolute paths or base64)."
+    },
+    "temperature": {
+      "type": "number",
+      "minimum": 0.0,
+      "maximum": 1.0,
+      "default": 0.2,
+      "description": "Response randomness. Examples: 0.1 (very focused), 0.2 (analytical), 0.5 (balanced)"
+    },
+    "reasoning_effort": {
+      "type": "string",
+      "enum": ["none", "minimal", "low", "medium", "high", "max"],
+      "default": "medium",
+      "description": "Reasoning depth for thinking models."
+    },
+    "use_websearch": {
+      "type": "boolean",
+      "default": false,
+      "description": "Enable web search for current information (models that support it)."
+    },
+    "async": {
+      "type": "boolean",
+      "default": false,
+      "description": "Execute the lap in background with per-turn progress tracking. Returns continuation_id immediately."
+    },
+    "export": {
+      "type": "boolean",
+      "default": false,
+      "description": "Export conversation to disk. Creates folder with continuation_id name containing numbered request/response files and metadata."
+    }
+  },
+  "required": ["prompt", "models"]
+}
+```
+
+#### Response Format
+
+**Synchronous Response (async=false):**
+
+The response content begins with a status line and `continuation_id:` line (the status line is omitted in the test environment), followed by a JSON result object:
+
+```
+✅ COMPLETED | CONVERSATION | conv_abc123 | 3.2s elapsed | 2/2 turns | codex, gemini
+continuation_id: conv_abc123
+
+{
+  "status": "conversation_complete",
+  "models_consulted": 2,
+  "successful_turns": 2,
+  "failed_turns": 0,
+  "turns": [
+    {
+      "model": "codex",
+      "provider": "codex",
+      "status": "success",
+      "response": "Opening analysis of the caching strategy...",
+      "position": 0
+    },
+    {
+      "model": "gemini",
+      "provider": "gemini-cli",
+      "status": "success",
+      "response": "Building on codex's point about TTLs, I'd add...",
+      "position": 1
+    }
+  ],
+  "continuation": {
+    "id": "conv_abc123",
+    "messageCount": 3
+  },
+  "settings": {
+    "temperature": 0.2,
+    "models_requested": ["codex", "gemini"]
+  }
+}
+```
+
+A turn that failed is recorded with `"status": "failed"` and an `"error"` note rather than aborting the lap; the response reports `successful_turns`/`models_consulted` accordingly and lists failed models in trailing failure details.
+
+**Asynchronous Response (async=true):**
+```json
+{
+  "content": "⏳ SUBMITTED | CONVERSATION | conv_xyz789 | 1/1 | Started: 01/12/2023 10:30:00 | \"Caching Round-Table\" | codex, gemini",
+  "continuation": {
+    "id": "conv_xyz789",
+    "status": "processing"
+  },
+  "async_execution": true
+}
+```
+
+When complete, `check_status` for the continuation_id renders the full lap transcript (the async result carries a top-level `content` field with the rendered transcript) plus the AI-generated title and final summary.
+
+#### Example Usage
+
+**Basic two-model lap:**
+```json
+{
+  "prompt": "Should we adopt event sourcing for the order service?",
+  "models": ["codex", "gemini"]
+}
+```
+
+**Continuing the round-table (another lap):**
+```json
+{
+  "prompt": "Now focus specifically on the migration path.",
+  "models": ["codex", "gemini"],
+  "continuation_id": "conv_abc123"
+}
+```
+
+**Async round-table with a custom per-turn instruction:**
+```json
+{
+  "prompt": "Review this module design.",
+  "models": ["codex", "gemini", "claude"],
+  "files": ["/project/src/orders/design.md"],
+  "turn_prompt": "Call out concrete failure modes you would test for.",
+  "async": true
+}
+```
+
 ## Supported Models
 
 ### OpenAI Models
@@ -1271,7 +1431,7 @@ describe('New Provider', () => {
 
 ### Overview
 
-Both Chat and Consensus tools support asynchronous execution mode for long-running operations. When `async: true` is specified:
+The Chat, Consensus, and Conversation tools support asynchronous execution mode for long-running operations. When `async: true` is specified:
 
 1. **Immediate Response**: Returns a `continuation_id` instantly
 2. **Background Processing**: Job runs in the background with streaming support

package/docs/EXAMPLES.md

@@ -520,6 +520,104 @@ CODEX_SANDBOX_MODE=danger-full-access
 }
 ```
 
+## 🔄 Conversation (Round-Table) Examples
+
+The `conversation` tool runs a turn-based round-table: models respond **in the order given**, and each model sees the full running transcript of every turn before it. One call = one lap. Pass the returned `continuation_id` to run another lap; every lap appends to one shared transcript. This differs from `consensus`, where all models answer the same prompt in parallel.
+
+### Basic Two-Model Round-Table
+
+```json
+{
+  "tool": "conversation",
+  "arguments": {
+    "prompt": "Should we adopt event sourcing for the order service?",
+    "models": ["codex", "gemini"]
+  }
+}
+```
+
+On this lap, `codex` opens, then `gemini` responds having seen codex's turn. The response contains both labeled turns in order plus a `continuation_id`.
+
+### Continuing the Round-Table (More Laps)
+
+```json
+// Lap 1 returns: "continuation": { "id": "conv_abc123" }
+
+// Lap 2 — every model again sees the full accumulated transcript
+{
+  "tool": "conversation",
+  "arguments": {
+    "prompt": "Now focus specifically on the migration path from the current design.",
+    "models": ["codex", "gemini"],
+    "continuation_id": "conv_abc123"
+  }
+}
+```
+
+You may also change the model list on a resuming lap (e.g. drop a participant or add one); the shared transcript persists regardless of who ran in earlier laps:
+
+```json
+{
+  "tool": "conversation",
+  "arguments": {
+    "prompt": "Bring in a third perspective on testability.",
+    "models": ["codex", "gemini", "claude"],
+    "continuation_id": "conv_abc123"
+  }
+}
+```
+
+### Round-Table with Files and a Custom Per-Turn Instruction
+
+```json
+{
+  "tool": "conversation",
+  "arguments": {
+    "prompt": "Review this module design and push back on weak assumptions.",
+    "models": ["codex", "gemini", "claude"],
+    "files": ["/c/Users/username/project/src/orders/design.md"],
+    "turn_prompt": "Call out concrete failure modes you would test for."
+  }
+}
+```
+
+### Async Round-Table with Progress Monitoring
+
+```json
+{
+  "tool": "conversation",
+  "arguments": {
+    "prompt": "Design a rollout plan for the new pricing engine.",
+    "models": ["codex", "gemini", "claude"],
+    "async": true
+  }
+}
+```
+
+**Immediate Response:**
+```json
+{
+  "content": "⏳ SUBMITTED | CONVERSATION | conv_xyz789 | 1/1 | Started: 01/12/2023 10:30:00 | \"Pricing Engine Rollout\" | codex, gemini, claude",
+  "continuation": {
+    "id": "conv_xyz789",
+    "status": "processing"
+  },
+  "async_execution": true
+}
+```
+
+**Monitor per-turn progress, then read the full transcript on completion:**
+```json
+{
+  "tool": "check_status",
+  "arguments": {
+    "continuation_id": "conv_xyz789"
+  }
+}
+```
+
+While running, the status line shows turn progress (e.g. `2/3 turns`) and the accumulating transcript. When complete, `check_status` renders the full lap transcript along with the AI-generated title and final summary.
+
 ## 🖼️ Image Analysis Examples
 
 ### Screenshot Analysis

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "converse-mcp-server",
-  "version": "2.22.8",
+  "version": "2.26.1",
   "description": "Converse MCP Server - Converse with other LLMs with chat and consensus tools",
   "type": "module",
   "main": "src/index.js",
@@ -94,30 +94,30 @@
     ".env.example"
   ],
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.126",
-    "@anthropic-ai/sdk": "^0.92.0",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.152",
+    "@anthropic-ai/sdk": "^0.99.0",
     "@github/copilot-sdk": "^0.3.0",
-    "@google/genai": "^1.51.0",
+    "@google/genai": "^2.7.0",
     "@mistralai/mistralai": "^2.2.1",
     "@modelcontextprotocol/sdk": "^1.29.0",
-    "@openai/codex-sdk": "^0.128.0",
-    "ai": "^6.0.174",
+    "@openai/codex-sdk": "^0.133.0",
+    "ai": "^6.0.191",
     "ai-sdk-provider-gemini-cli": "^2.0.1",
     "cors": "^2.8.6",
     "dotenv": "^17.4.2",
     "express": "^5.2.1",
-    "lru-cache": "^11.3.5",
+    "lru-cache": "^11.5.0",
     "nanoid": "^5.1.11",
-    "openai": "^6.35.0",
+    "openai": "^6.39.0",
     "p-limit": "^7.3.0",
-    "vite": "^8.0.10"
+    "vite": "^8.0.14"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^4.1.5",
+    "@vitest/coverage-v8": "^4.1.7",
     "cross-env": "^10.1.0",
-    "eslint": "^10.3.0",
+    "eslint": "^10.4.0",
     "prettier": "^3.8.3",
     "rimraf": "^6.1.3",
-    "vitest": "^4.1.5"
+    "vitest": "^4.1.7"
   }
 }

package/src/async/asyncJobStore.js

@@ -167,9 +167,9 @@ class LRUAsyncJobStore extends AsyncJobStoreInterface {
   async create(tool, options = {}) {
     try {
       // Validate parameters
-      if (!tool || !['chat', 'consensus'].includes(tool)) {
+      if (!tool || !['chat', 'consensus', 'conversation'].includes(tool)) {
         throw new AsyncJobStoreError(
-          'Invalid tool: must be "chat" or "consensus"',
+          'Invalid tool: must be "chat", "consensus", or "conversation"',
           'INVALID_TOOL',
         );
       }

package/src/providers/anthropic.js

@@ -13,6 +13,39 @@ import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 
 // Define supported Claude models with their capabilities
 const SUPPORTED_MODELS = {
+  'claude-opus-4-8': {
+    modelName: 'claude-opus-4-8',
+    friendlyName: 'Claude Opus 4.8',
+    contextWindow: 200000,
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: true,
+    supportsWebSearch: false,
+    supportsThinking: true,
+    supportsAdaptiveThinking: true,
+    minThinkingTokens: 1024,
+    maxThinkingTokens: 128000,
+    timeout: 600000,
+    supportsEffort: true,
+    effortGA: true,
+    supports1MContext: true,
+    supportsCompaction: true,
+    description:
+      'Claude Opus 4.8 - Most capable model for complex reasoning and agentic coding',
+    aliases: [
+      'claude-opus-4-8',
+      'claude-4.8-opus',
+      'claude-4-8-opus',
+      'opus-4.8',
+      'opus-4-8',
+      'opus4.8',
+      'opus4-8',
+      'claude-opus-4.8',
+      'opus',
+      'claude-opus',
+    ],
+  },
   'claude-opus-4-7': {
     modelName: 'claude-opus-4-7',
     friendlyName: 'Claude Opus 4.7',
@@ -32,7 +65,7 @@ const SUPPORTED_MODELS = {
     supports1MContext: true,
     supportsCompaction: true,
     description:
-      'Claude Opus 4.7 - Most capable model for complex reasoning and agentic coding',
+      'Claude Opus 4.7 - Previous most capable model for complex reasoning and agentic coding',
     aliases: [
       'claude-opus-4-7',
       'claude-4.7-opus',
@@ -42,8 +75,6 @@ const SUPPORTED_MODELS = {
       'opus4.7',
       'opus4-7',
       'claude-opus-4.7',
-      'opus',
-      'claude-opus',
     ],
   },
   'claude-opus-4-6': {
@@ -243,7 +274,7 @@ const THINKING_BUDGETS = {
 };
 
 /**
- * Effort parameter mapping for Opus 4.7, Opus 4.6, Sonnet 4.6, and Opus 4.5
+ * Effort parameter mapping for Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 4.6, and Opus 4.5
  * Maps reasoning_effort values to Anthropic's effort parameter values
  */
 const EFFORT_MAP = {

package/src/providers/claude.js

@@ -223,7 +223,7 @@ async function* createStreamingGenerator(
     // Build query options
     // Use higher maxTurns to allow for file reading operations
     const queryOptions = {
-      model: 'claude-opus-4-7', // Use Opus 4.7 for best quality
+      model: 'claude-opus-4-8', // Use Opus 4.8 for best quality
       maxTurns: 20, // Allow multiple turns for file operations
       permissionMode: 'bypassPermissions', // Don't prompt for permissions
     };

package/src/providers/copilot.js

@@ -249,7 +249,20 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: false,
     timeout: 600000,
     description: 'Anthropic Claude Opus 4.7 via Copilot subscription',
-    aliases: ['opus'],
+    aliases: [],
+  },
+  'claude-opus-4.8': {
+    modelName: 'claude-opus-4.8',
+    friendlyName: 'Claude Opus 4.8 (via Copilot)',
+    contextWindow: 200000,
+    maxOutputTokens: 32768,
+    supportsStreaming: true,
+    supportsImages: false,
+    supportsTemperature: false,
+    supportsWebSearch: false,
+    timeout: 600000,
+    description: 'Anthropic Claude Opus 4.8 via Copilot subscription',
+    aliases: ['opus', 'claude'],
   },
 
   // Google models

package/src/providers/gemini-cli.js

@@ -26,6 +26,32 @@ import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 const SUPPORTED_MODELS = {
   gemini: {
     modelName: 'gemini',
+    friendlyName: 'Gemini 3.5 Flash (via CLI)',
+    contextWindow: 1048576, // 1M tokens
+    maxOutputTokens: 65536,
+    supportsStreaming: true,
+    supportsImages: true, // Base64 only (no URLs)
+    supportsTemperature: true,
+    supportsThinking: true,
+    supportsWebSearch: true,
+    timeout: 600000, // 10 minutes
+    description:
+      'Gemini 3.5 Flash via OAuth - frontier agentic/coding at Flash speed (requires Gemini CLI authentication)',
+    aliases: [
+      'gemini-cli',
+      'gemini-3.5-flash',
+      'gemini-3.5',
+      'gemini3.5',
+      'flash',
+      'flash-3.5',
+      'gemini-flash',
+      'gemini-flash-3.5',
+    ],
+    // Internal SDK model name passed to the Google Cloud Code endpoint
+    sdkModelName: 'gemini-3.5-flash',
+  },
+  'gemini-3.1-pro-preview': {
+    modelName: 'gemini-3.1-pro-preview',
     friendlyName: 'Gemini 3.1 Pro Preview (via CLI)',
     contextWindow: 1048576, // 1M tokens
     maxOutputTokens: 64000,
@@ -37,8 +63,13 @@ const SUPPORTED_MODELS = {
     timeout: 600000, // 10 minutes
     description:
       'Gemini 3.1 Pro Preview via OAuth - requires Gemini CLI authentication',
-    aliases: ['gemini-cli'],
-    // Internal SDK model name (user-facing "gemini" maps to SDK's "gemini-3.1-pro-preview")
+    aliases: [
+      'gemini-3.1-pro',
+      'gemini-3.1',
+      'gemini-pro',
+      'gemini-3-pro',
+      'pro',
+    ],
     sdkModelName: 'gemini-3.1-pro-preview',
   },
 };

package/src/providers/google.js

@@ -106,6 +106,32 @@ const SUPPORTED_MODELS = {
       'pro',
     ],
   },
+  'gemini-3.5-flash': {
+    modelName: 'gemini-3.5-flash',
+    friendlyName: 'Gemini (Flash 3.5)',
+    contextWindow: 1048576, // 1M tokens
+    maxOutputTokens: 65536,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: true,
+    supportsThinking: true,
+    supportsWebSearch: true,
+    thinkingMode: 'level',
+    thinkingLevels: ['minimal', 'low', 'medium', 'high'],
+    timeout: 300000,
+    description:
+      'Gemini 3.5 Flash - Frontier-level agentic and coding performance at Flash speed (1M context)',
+    aliases: [
+      'gemini-3.5',
+      'gemini3.5',
+      'gemini-3.5-flash-latest',
+      'flash-3.5',
+      'flash3.5',
+      'gemini-flash-3.5',
+      'gemini flash 3.5',
+      '3.5-flash',
+    ],
+  },
 };
 
 // Thinking mode budget percentages

package/src/systemPrompts.js

@@ -88,3 +88,36 @@ QUALITY STANDARDS
 
 Remember: The best solution often has one breakthrough insight that makes the complexity fall away.
 `.trim();
+
+/**
+ * Conversation tool system prompt - sequential round-table dialogue.
+ *
+ * Unlike CONSENSUS_PROMPT (parallel, rigidly-structured answers), this frames a
+ * turn-based round-table where each participant speaks after seeing the full
+ * running transcript and is expected to build the discussion forward as dialogue.
+ */
+export const CONVERSATION_PROMPT = `
+You are taking part in a multi-model round-table conversation. Several AI models speak one after another, each seeing the full running transcript of everything said before it. When it is your turn, you respond to the whole conversation so far and your response is passed on to the next participant.
+
+Your goal: advance the discussion. Build on, challenge, or refine what earlier participants have said — don't merely repeat them. Add genuine value with each turn, whether that's a new insight, a correction, a synthesis, or a sharper framing. Treat this as a collaborative dialogue, not a set of isolated answers.
+
+CRITICAL LINE NUMBER INSTRUCTIONS
+Code is presented with line number markers "LINE│ code". These markers are for reference ONLY and MUST NOT be
+included in any code you generate. Always reference specific line numbers in your replies in order to locate
+exact positions if needed to point to exact locations. Include a very short code excerpt alongside for clarity.
+Never include "LINE│" markers in generated code snippets.
+
+IF MORE INFORMATION IS NEEDED
+If you need to see specific code, files, or technical context to properly contribute to the discussion, respond with this exact JSON:
+{
+  "status": "files_required_to_continue",
+  "mandatory_instructions": "<your critical instructions for the agent>",
+  "files_needed": ["[file name here]", "[or some folder/]"]
+}
+
+RESPONSE STYLE
+- Respond as a dialogue turn, not a rigid report — speak naturally to the round-table.
+- Reference earlier participants by name when you agree, disagree, or extend their points.
+- Be direct and technical; surface trade-offs and challenge weak assumptions constructively.
+- Keep momentum: leave the conversation in a better place than you found it for the next participant.
+`.trim();