ai-cli-mcp 2.0.0 → 2.1.0

package/.claude/settings.local.json

@@ -12,7 +12,8 @@
       "mcp__ccm__get_claude_result",
       "mcp__chat__agent_communication_list_rooms",
       "mcp__chat__agent_communication_get_messages",
-      "mcp__ccm__list_claude_processes"
+      "mcp__ccm__list_claude_processes",
+      "Bash(gemini:*)"
     ],
     "deny": []
   }

package/README.md

@@ -1,13 +1,11 @@
 # AI CLI MCP Server
 
-<img src="assets/claude_code_mcp_logo.png" alt="AI CLI MCP Logo">
-
 [![npm package](https://img.shields.io/npm/v/ai-cli-mcp)](https://www.npmjs.com/package/ai-cli-mcp)
 [![View changelog](https://img.shields.io/badge/Explore%20Changelog-brightgreen)](/CHANGELOG.md)
 
 > **📦 Package Migration Notice**: This package was formerly `@mkxultra/claude-code-mcp` and has been renamed to `ai-cli-mcp` to reflect its expanded support for multiple AI CLI tools.
 
-An MCP (Model Context Protocol) server that allows running AI CLI tools (Claude and Codex) in background processes with automatic permission handling.
+An MCP (Model Context Protocol) server that allows running AI CLI tools (Claude, Codex, and Gemini) in background processes with automatic permission handling.
 
 Did you notice that Cursor sometimes struggles with complex, multi-step edits or operations? This server, with its powerful unified `run` tool, enables multiple AI agents to handle your coding tasks more effectively.
 
@@ -19,7 +17,8 @@ This MCP server provides tools that can be used by LLMs to interact with AI CLI
 
 - Run Claude CLI with all permissions bypassed (using `--dangerously-skip-permissions`)
 - Execute Codex CLI with automatic approval mode (using `--full-auto`)
-- Support multiple AI models: Claude (sonnet, opus, haiku) and Codex (gpt-5-low, gpt-5-medium, gpt-5-high)
+- Execute Gemini CLI with automatic approval mode (using `-y`)
+- Support multiple AI models: Claude (sonnet, opus, haiku), Codex (gpt-5-low, gpt-5-medium, gpt-5-high), and Gemini (gemini-2.5-pro, gemini-2.5-flash)
 - Manage background processes with PID tracking
 - Parse and return structured outputs from both tools
 
@@ -31,13 +30,12 @@ This MCP server provides tools that can be used by LLMs to interact with AI CLI
 - Claude has wider system access and can do things that Cursor/Windsurf can't do (or believe they can't), so whenever they are stuck just ask them "use claude code" and it will usually un-stuck them.
 - Agents in Agents rules.
 
-<img src="assets/agents_in_agents_meme.jpg" alt="Agents in Agents Meme">
-
 ## Prerequisites
 
 - Node.js v20 or later (Use fnm or nvm to install)
 - Claude CLI installed locally (run it and call /doctor) and `--dangerously-skip-permissions` accepted
 - Codex CLI installed (optional, for Codex support)
+- Gemini CLI installed (optional, for Gemini support)
 
 ## Configuration
 
@@ -45,9 +43,10 @@ This MCP server provides tools that can be used by LLMs to interact with AI CLI
 
 - `CLAUDE_CLI_NAME`: Override the Claude CLI binary name or provide an absolute path (default: `claude`)
 - `CODEX_CLI_NAME`: Override the Codex CLI binary name or provide an absolute path (default: `codex`)
+- `GEMINI_CLI_NAME`: Override the Gemini CLI binary name or provide an absolute path (default: `gemini`)
 - `MCP_CLAUDE_DEBUG`: Enable debug logging (set to `true` for verbose output)
 
-Both CLI name variables support:
+All CLI name variables support:
 - Simple name: `CLAUDE_CLI_NAME=claude-custom` or `CODEX_CLI_NAME=codex-v2`
 - Absolute path: `CLAUDE_CLI_NAME=/path/to/custom/claude`
 
@@ -112,7 +111,15 @@ Follow the prompts to accept. Once this is done, the MCP server will be able to
 codex login
 ```
 
-macOS might ask for folder permissions the first time either tool runs. If the first run fails, subsequent runs should work.
+### For Gemini CLI:
+
+**For Gemini, ensure you're logged in and have configured your credentials:**
+
+```bash
+gemini auth login
+```
+
+macOS might ask for folder permissions the first time any of these tools run. If the first run fails, subsequent runs should work.
 
 ## Connecting to Your MCP Client
 
@@ -155,6 +162,7 @@ Executes a prompt using either Claude CLI or Codex CLI. The appropriate CLI is a
 - `model` (string, optional): The model to use:
   - Claude models: "sonnet", "opus", "haiku"
   - Codex models: "gpt-5-low", "gpt-5-medium", "gpt-5-high"
+  - Gemini models: "gemini-2.5-pro", "gemini-2.5-flash"
 - `session_id` (string, optional): Optional session ID to resume a previous session. Supported for: haiku, sonnet, opus.
 
 ### `list_processes`
@@ -199,6 +207,18 @@ Terminates a running AI agent process by PID.
 }
 ```
 
+**Example with Gemini:**
+```json
+{
+  "toolName": "run",
+  "arguments": {
+    "prompt": "Generate unit tests for the Calculator class",
+    "workFolder": "/Users/username/my_project",
+    "model": "gemini-2.5-pro"
+  }
+}
+```
+
 ### Examples
 
 Here are some visual examples of the server in action:
@@ -308,6 +328,31 @@ npm run test:coverage
 
 For detailed testing documentation, see our [E2E Testing Guide](./docs/e2e-testing.md).
 
+## Manual Testing with MCP Inspector
+
+You can manually test the MCP server using the Model Context Protocol Inspector:
+
+```bash
+# Build the project first
+npm run build
+
+# Start the MCP Inspector with the server
+npx @modelcontextprotocol/inspector node dist/server.js
+```
+
+This will open a web interface where you can:
+1. View all available tools (`run`, `list_processes`, `get_result`, `kill_process`)
+2. Test each tool with different parameters
+3. Test different AI models including:
+   - Claude models: `sonnet`, `opus`, `haiku`
+   - Codex models: `gpt-5-low`, `gpt-5-medium`, `gpt-5-high`
+   - Gemini models: `gemini-2.5-pro`, `gemini-2.5-flash`
+
+Example test: Select the `run` tool and provide:
+- `prompt`: "What is 2+2?"
+- `workFolder`: "/tmp"
+- `model`: "gemini-2.5-flash"
+
 ## Configuration via Environment Variables
 
 The server's behavior can be customized using these environment variables:

package/dist/parsers.js

@@ -52,3 +52,17 @@ export function parseClaudeOutput(stdout) {
         return null;
     }
 }
+/**
+ * Parse Gemini JSON output
+ */
+export function parseGeminiOutput(stdout) {
+    if (!stdout)
+        return null;
+    try {
+        return JSON.parse(stdout);
+    }
+    catch (e) {
+        debugLog(`[Debug] Failed to parse Gemini JSON output: ${e}`);
+        return null;
+    }
+}

package/dist/server.js

@@ -7,9 +7,9 @@ import { existsSync, readFileSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { join, resolve as pathResolve } from 'node:path';
 import * as path from 'path';
-import { parseCodexOutput, parseClaudeOutput } from './parsers.js';
+import { parseCodexOutput, parseClaudeOutput, parseGeminiOutput } from './parsers.js';
 // Server version - update this when releasing new versions
-const SERVER_VERSION = "2.0.0";
+const SERVER_VERSION = "2.0.1";
 // Model alias mappings for user-friendly model names
 const MODEL_ALIASES = {
     'haiku': 'claude-3-5-haiku-20241022'
@@ -28,6 +28,42 @@ export function debugLog(message, ...optionalParams) {
         console.error(message, ...optionalParams);
     }
 }
+/**
+ * Determine the Gemini CLI command/path.
+ * Similar to findClaudeCli but for Gemini
+ */
+export function findGeminiCli() {
+    debugLog('[Debug] Attempting to find Gemini CLI...');
+    // Check for custom CLI name from environment variable
+    const customCliName = process.env.GEMINI_CLI_NAME;
+    if (customCliName) {
+        debugLog(`[Debug] Using custom Gemini CLI name from GEMINI_CLI_NAME: ${customCliName}`);
+        // If it's an absolute path, use it directly
+        if (path.isAbsolute(customCliName)) {
+            debugLog(`[Debug] GEMINI_CLI_NAME is an absolute path: ${customCliName}`);
+            return customCliName;
+        }
+        // If it starts with ~ or ./, reject as relative paths are not allowed
+        if (customCliName.startsWith('./') || customCliName.startsWith('../') || customCliName.includes('/')) {
+            throw new Error(`Invalid GEMINI_CLI_NAME: Relative paths are not allowed. Use either a simple name (e.g., 'gemini') or an absolute path (e.g., '/tmp/gemini-test')`);
+        }
+    }
+    const cliName = customCliName || 'gemini';
+    // Try local install path: ~/.gemini/local/gemini
+    const userPath = join(homedir(), '.gemini', 'local', 'gemini');
+    debugLog(`[Debug] Checking for Gemini CLI at local user path: ${userPath}`);
+    if (existsSync(userPath)) {
+        debugLog(`[Debug] Found Gemini CLI at local user path: ${userPath}. Using this path.`);
+        return userPath;
+    }
+    else {
+        debugLog(`[Debug] Gemini CLI not found at local user path: ${userPath}.`);
+    }
+    // Fallback to CLI name (PATH lookup)
+    debugLog(`[Debug] Falling back to "${cliName}" command name, relying on spawn/PATH lookup.`);
+    console.warn(`[Warning] Gemini CLI not found at ~/.gemini/local/gemini. Falling back to "${cliName}" in PATH. Ensure it is installed and accessible.`);
+    return cliName;
+}
 /**
  * Determine the Codex CLI command/path.
  * Similar to findClaudeCli but for Codex
@@ -163,14 +199,17 @@ export class ClaudeCodeServer {
     server;
     claudeCliPath;
     codexCliPath;
+    geminiCliPath;
     sigintHandler;
     packageVersion;
     constructor() {
         // Use the simplified findClaudeCli function
         this.claudeCliPath = findClaudeCli(); // Removed debugMode argument
         this.codexCliPath = findCodexCli();
+        this.geminiCliPath = findGeminiCli();
         console.error(`[Setup] Using Claude CLI command/path: ${this.claudeCliPath}`);
         console.error(`[Setup] Using Codex CLI command/path: ${this.codexCliPath}`);
+        console.error(`[Setup] Using Gemini CLI command/path: ${this.geminiCliPath}`);
         this.packageVersion = SERVER_VERSION;
         this.server = new Server({
             name: 'ai_cli_mcp',
@@ -197,7 +236,7 @@ export class ClaudeCodeServer {
             tools: [
                 {
                     name: 'run',
-                    description: `AI Agent Runner: Starts a Claude or Codex CLI process in the background and returns a PID immediately. Use list_processes and get_result to monitor progress.
+                    description: `AI Agent Runner: Starts a Claude, Codex, or Gemini CLI process in the background and returns a PID immediately. Use list_processes and get_result to monitor progress.
 
 • File ops: Create, read, (fuzzy) edit, move, copy, delete, list files, analyze/ocr images, file content analysis
 • Code: Generate / analyse / refactor / fix
@@ -208,8 +247,8 @@ export class ClaudeCodeServer {
 
 **IMPORTANT**: This tool now returns immediately with a PID. Use other tools to check status and get results.
 
-**Supported models**: 
-"sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high"
+**Supported models**:
+"sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high", "gemini-2.5-pro", "gemini-2.5-flash"
 
 **Prompt input**: You must provide EITHER prompt (string) OR prompt_file (file path), but not both.
 
@@ -223,11 +262,6 @@ export class ClaudeCodeServer {
                     inputSchema: {
                         type: 'object',
                         properties: {
-                            agent: {
-                                type: 'string',
-                                description: 'The agent to use: "claude" or "codex". Defaults to "claude".',
-                                enum: ['claude', 'codex'],
-                            },
                             prompt: {
                                 type: 'string',
                                 description: 'The detailed natural language prompt for the agent to execute. Either this or prompt_file is required.',
@@ -242,7 +276,7 @@ export class ClaudeCodeServer {
                             },
                             model: {
                                 type: 'string',
-                                description: 'The model to use: "sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high".',
+                                description: 'The model to use: "sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high", "gemini-2.5-pro", "gemini-2.5-flash".',
                             },
                             session_id: {
                                 type: 'string',
@@ -360,7 +394,16 @@ export class ClaudeCodeServer {
         }
         // Determine which agent to use based on model name
         const model = toolArguments.model || '';
-        const agent = model.startsWith('gpt-') ? 'codex' : 'claude';
+        let agent;
+        if (model.startsWith('gpt-')) {
+            agent = 'codex';
+        }
+        else if (model.startsWith('gemini')) {
+            agent = 'gemini';
+        }
+        else {
+            agent = 'claude';
+        }
         let cliPath;
         let processArgs;
         if (agent === 'codex') {
@@ -378,6 +421,17 @@ export class ClaudeCodeServer {
             }
             processArgs.push('--full-auto', '--json', prompt);
         }
+        else if (agent === 'gemini') {
+            // Handle Gemini
+            cliPath = this.geminiCliPath;
+            processArgs = ['-y', '--output-format', 'json'];
+            // Add model if specified
+            if (toolArguments.model) {
+                processArgs.push('--model', toolArguments.model);
+            }
+            // Add prompt as positional argument
+            processArgs.push(prompt);
+        }
         else {
             // Handle Claude (default)
             cliPath = this.claudeCliPath;
@@ -515,6 +569,9 @@ export class ClaudeCodeServer {
             else if (process.toolType === 'claude') {
                 agentOutput = parseClaudeOutput(process.stdout);
             }
+            else if (process.toolType === 'gemini') {
+                agentOutput = parseGeminiOutput(process.stdout);
+            }
         }
         // Construct response with agent's output and process metadata
         const response = {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "ai-cli-mcp",
-  "version": "2.0.0",
-  "description": "MCP server for AI CLI tools (Claude and Codex) with background process management",
+  "version": "2.1.0",
+  "description": "MCP server for AI CLI tools (Claude, Codex, and Gemini) with background process management",
   "author": "mkXultra",
   "license": "MIT",
   "main": "dist/server.js",

package/src/parsers.ts

@@ -45,11 +45,25 @@ export function parseCodexOutput(stdout: string): any {
  */
 export function parseClaudeOutput(stdout: string): any {
   if (!stdout) return null;
-  
+
   try {
     return JSON.parse(stdout);
   } catch (e) {
     debugLog(`[Debug] Failed to parse Claude JSON output: ${e}`);
     return null;
   }
+}
+
+/**
+ * Parse Gemini JSON output
+ */
+export function parseGeminiOutput(stdout: string): any {
+  if (!stdout) return null;
+
+  try {
+    return JSON.parse(stdout);
+  } catch (e) {
+    debugLog(`[Debug] Failed to parse Gemini JSON output: ${e}`);
+    return null;
+  }
 }

package/src/server.ts

@@ -13,10 +13,10 @@ import { existsSync, readFileSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { join, resolve as pathResolve } from 'node:path';
 import * as path from 'path';
-import { parseCodexOutput, parseClaudeOutput } from './parsers.js';
+import { parseCodexOutput, parseClaudeOutput, parseGeminiOutput } from './parsers.js';
 
 // Server version - update this when releasing new versions
-const SERVER_VERSION = "2.0.0";
+const SERVER_VERSION = "2.1.0";
 
 // Model alias mappings for user-friendly model names
 const MODEL_ALIASES: Record<string, string> = {
@@ -39,7 +39,7 @@ interface ClaudeProcess {
   prompt: string;
   workFolder: string;
   model?: string;
-  toolType: 'claude' | 'codex';  // Identify which CLI tool
+  toolType: 'claude' | 'codex' | 'gemini';  // Identify which CLI tool
   startTime: string;
   stdout: string;
   stderr: string;
@@ -57,6 +57,49 @@ export function debugLog(message?: any, ...optionalParams: any[]): void {
   }
 }
 
+/**
+ * Determine the Gemini CLI command/path.
+ * Similar to findClaudeCli but for Gemini
+ */
+export function findGeminiCli(): string {
+  debugLog('[Debug] Attempting to find Gemini CLI...');
+
+  // Check for custom CLI name from environment variable
+  const customCliName = process.env.GEMINI_CLI_NAME;
+  if (customCliName) {
+    debugLog(`[Debug] Using custom Gemini CLI name from GEMINI_CLI_NAME: ${customCliName}`);
+
+    // If it's an absolute path, use it directly
+    if (path.isAbsolute(customCliName)) {
+      debugLog(`[Debug] GEMINI_CLI_NAME is an absolute path: ${customCliName}`);
+      return customCliName;
+    }
+
+    // If it starts with ~ or ./, reject as relative paths are not allowed
+    if (customCliName.startsWith('./') || customCliName.startsWith('../') || customCliName.includes('/')) {
+      throw new Error(`Invalid GEMINI_CLI_NAME: Relative paths are not allowed. Use either a simple name (e.g., 'gemini') or an absolute path (e.g., '/tmp/gemini-test')`);
+    }
+  }
+
+  const cliName = customCliName || 'gemini';
+
+  // Try local install path: ~/.gemini/local/gemini
+  const userPath = join(homedir(), '.gemini', 'local', 'gemini');
+  debugLog(`[Debug] Checking for Gemini CLI at local user path: ${userPath}`);
+
+  if (existsSync(userPath)) {
+    debugLog(`[Debug] Found Gemini CLI at local user path: ${userPath}. Using this path.`);
+    return userPath;
+  } else {
+    debugLog(`[Debug] Gemini CLI not found at local user path: ${userPath}.`);
+  }
+
+  // Fallback to CLI name (PATH lookup)
+  debugLog(`[Debug] Falling back to "${cliName}" command name, relying on spawn/PATH lookup.`);
+  console.warn(`[Warning] Gemini CLI not found at ~/.gemini/local/gemini. Falling back to "${cliName}" in PATH. Ensure it is installed and accessible.`);
+  return cliName;
+}
+
 /**
  * Determine the Codex CLI command/path.
  * Similar to findClaudeCli but for Codex
@@ -232,6 +275,7 @@ export class ClaudeCodeServer {
   private server: Server;
   private claudeCliPath: string;
   private codexCliPath: string;
+  private geminiCliPath: string;
   private sigintHandler?: () => Promise<void>;
   private packageVersion: string;
 
@@ -239,8 +283,10 @@ export class ClaudeCodeServer {
     // Use the simplified findClaudeCli function
     this.claudeCliPath = findClaudeCli(); // Removed debugMode argument
     this.codexCliPath = findCodexCli();
+    this.geminiCliPath = findGeminiCli();
     console.error(`[Setup] Using Claude CLI command/path: ${this.claudeCliPath}`);
     console.error(`[Setup] Using Codex CLI command/path: ${this.codexCliPath}`);
+    console.error(`[Setup] Using Gemini CLI command/path: ${this.geminiCliPath}`);
     this.packageVersion = SERVER_VERSION;
 
     this.server = new Server(
@@ -274,7 +320,7 @@ export class ClaudeCodeServer {
       tools: [
         {
           name: 'run',
-          description: `AI Agent Runner: Starts a Claude or Codex CLI process in the background and returns a PID immediately. Use list_processes and get_result to monitor progress.
+          description: `AI Agent Runner: Starts a Claude, Codex, or Gemini CLI process in the background and returns a PID immediately. Use list_processes and get_result to monitor progress.
 
 • File ops: Create, read, (fuzzy) edit, move, copy, delete, list files, analyze/ocr images, file content analysis
 • Code: Generate / analyse / refactor / fix
@@ -285,8 +331,8 @@ export class ClaudeCodeServer {
 
 **IMPORTANT**: This tool now returns immediately with a PID. Use other tools to check status and get results.
 
-**Supported models**: 
-"sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high"
+**Supported models**:
+"sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high", "gemini-2.5-pro", "gemini-2.5-flash"
 
 **Prompt input**: You must provide EITHER prompt (string) OR prompt_file (file path), but not both.
 
@@ -300,11 +346,6 @@ export class ClaudeCodeServer {
           inputSchema: {
             type: 'object',
             properties: {
-              agent: {
-                type: 'string',
-                description: 'The agent to use: "claude" or "codex". Defaults to "claude".',
-                enum: ['claude', 'codex'],
-              },
               prompt: {
                 type: 'string',
                 description: 'The detailed natural language prompt for the agent to execute. Either this or prompt_file is required.',
@@ -319,7 +360,7 @@ export class ClaudeCodeServer {
               },
               model: {
                 type: 'string',
-                description: 'The model to use: "sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high".',
+                description: 'The model to use: "sonnet", "opus", "haiku", "gpt-5-low", "gpt-5-medium", "gpt-5-high", "gemini-2.5-pro", "gemini-2.5-flash".',
               },
               session_id: {
                 type: 'string',
@@ -449,16 +490,24 @@ export class ClaudeCodeServer {
 
     // Determine which agent to use based on model name
     const model = toolArguments.model || '';
-    const agent = model.startsWith('gpt-') ? 'codex' : 'claude';
-    
+    let agent: 'codex' | 'claude' | 'gemini';
+
+    if (model.startsWith('gpt-')) {
+      agent = 'codex';
+    } else if (model.startsWith('gemini')) {
+      agent = 'gemini';
+    } else {
+      agent = 'claude';
+    }
+
     let cliPath: string;
     let processArgs: string[];
-    
+
     if (agent === 'codex') {
       // Handle Codex
       cliPath = this.codexCliPath;
       processArgs = ['exec'];
-      
+
       // Parse model format for Codex (e.g., gpt-5-low -> model: gpt-5, effort: low)
       if (toolArguments.model) {
         // Split by "gpt-5-" to get the effort level
@@ -468,19 +517,32 @@ export class ClaudeCodeServer {
         }
         processArgs.push('--model', 'gpt-5');
       }
-      
+
       processArgs.push('--full-auto', '--json', prompt);
-      
+
+    } else if (agent === 'gemini') {
+      // Handle Gemini
+      cliPath = this.geminiCliPath;
+      processArgs = ['-y', '--output-format', 'json'];
+
+      // Add model if specified
+      if (toolArguments.model) {
+        processArgs.push('--model', toolArguments.model);
+      }
+
+      // Add prompt as positional argument
+      processArgs.push(prompt);
+
     } else {
       // Handle Claude (default)
       cliPath = this.claudeCliPath;
       processArgs = ['--dangerously-skip-permissions', '--output-format', 'json'];
-      
+
       // Add session_id if provided (Claude only)
       if (toolArguments.session_id && typeof toolArguments.session_id === 'string') {
         processArgs.push('-r', toolArguments.session_id);
       }
-      
+
       processArgs.push('-p', prompt);
       if (toolArguments.model && typeof toolArguments.model === 'string') {
         const resolvedModel = resolveModelAlias(toolArguments.model);
@@ -507,7 +569,7 @@ export class ClaudeCodeServer {
       prompt,
       workFolder: effectiveCwd,
       model: toolArguments.model,
-      toolType: agent as 'claude' | 'codex',
+      toolType: agent,
       startTime: new Date().toISOString(),
       stdout: '',
       stderr: '',
@@ -625,6 +687,8 @@ export class ClaudeCodeServer {
         agentOutput = parseCodexOutput(process.stdout);
       } else if (process.toolType === 'claude') {
         agentOutput = parseClaudeOutput(process.stdout);
+      } else if (process.toolType === 'gemini') {
+        agentOutput = parseGeminiOutput(process.stdout);
       }
     }