@xagent-ai/cli 1.2.2 → 1.3.0

package/docs/architecture/mcp-integration-guide.md

@@ -1,194 +1,304 @@
-# SKILL, Workflow and MCP Integration Guide
-
-This document explains how to integrate external skills, workflows, and MCP servers with xAgent CLI.
-
-## Overview
-
-xAgent CLI supports extensibility through:
-- **SKILL**: Reusable task modules
-- **Workflow**: Predefined automation sequences
-- **MCP (Model Context Protocol)**: External tool servers
-
-## MCP Integration
-
-### What is MCP?
-
-MCP (Model Context Protocol) is an open standard that enables AI assistants to connect with external tools and data sources.
-
-### Configuring MCP Servers
-
-Add MCP servers to your settings file (`~/.xagent/settings.json`):
-
-```json
-{
-  "mcpServers": {
-    "filesystem": {
-      "command": "npx",
-      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/directory"]
-    },
-    "github": {
-      "command": "npx",
-      "args": ["-y", "@modelcontextprotocol/server-github"]
-    }
-  }
-}
-```
-
-### MCP Server Configuration
-
-```typescript
-interface MCPServerConfig {
-  command?: string;           // Command for stdio transport
-  args?: string[];            // Command arguments
-  env?: Record<string, string>;  // Environment variables
-  url?: string;               // URL for HTTP/SSE transport
-  transport?: 'stdio' | 'sse' | 'http';
-  authToken?: string;
-  headers?: Record<string, string>;
-  timeout?: number;
-}
-```
-
-### Supported MCP Servers
-
-| Server | Description |
-|--------|-------------|
-| filesystem | File system operations |
-| github | GitHub API integration |
-| postgres | PostgreSQL database |
-| redis | Redis key-value store |
-| puppeteer | Browser automation |
-
-### MCP Tool Synchronization
-
-MCP tools are automatically registered with the ToolRegistry and available to agents. MCP tool names use a suffix pattern to avoid conflicts:
-
-```
-{original_name}_mcp{server_number}
-```
-
-For example: `github__create_issue_mcp0`
-
-## Remote Mode MCP Sync
-
-When using Remote Mode (OAuth authentication), MCP tools are synchronized to the remote server:
-
-1. MCP server connects and discovers tools
-2. Tool definitions are sent to remote server
-3. Remote LLM can call MCP tools through the remote client
-4. Tool execution results are synced back
-
-## Workflow System
-
-### What are Workflows?
-
-Workflows are predefined sequences of actions that automate complex tasks. They can be installed from the Workflow Market.
-
-### Installing Workflows
-
-```bash
-xagent workflow --add <workflow-id>
-```
-
-### Workflow Structure
-
-```typescript
-interface WorkflowConfig {
-  id: string;
-  name: string;
-  description: string;
-  version: string;
-  author: string;
-  agents: AgentConfig[];
-  commands: Record<string, string>;
-  mcpServers: Record<string, MCPServerConfig>;
-  xagentMd: string;
-  files: Record<string, string>;
-}
-```
-
-### Managing Workflows
-
-```bash
-# List installed workflows
-xagent workflow --list
-
-# Remove a workflow
-xagent workflow --remove <workflow-id>
-```
-
-## SKILL System
-
-### What are SKILLs?
-
-SKILLs are reusable task modules that can be composed to create complex automation flows. SKILLs are invoked through the `InvokeSkill` tool.
-
-### SKILL Structure
-
-```typescript
-interface SkillConfig {
-  id: string;
-  name: string;
-  description: string;
-  parameters: Record<string, SkillParameter>;
-  execute: (input: Record<string, unknown>) => Promise<SkillResult>;
-}
-```
-
-### Available Skills
-
-xAgent includes several built-in skills:
-
-| Skill | Category | Description |
-|-------|----------|-------------|
-| docx | Document Processing | Create and edit Word documents |
-| pptx | Document Processing | Create PowerPoint presentations |
-| xlsx | Document Processing | Create and edit Excel spreadsheets |
-| pdf | Document Processing | PDF processing and manipulation |
-| mcp-builder | Development | Build custom MCP servers |
-| skill-creator | Development | Create new skills |
-| webapp-testing | Testing | Web application testing |
-| frontend-design | Design | Frontend design and implementation |
-| theme-factory | Design | Color theme generation |
-| algorithmic-art | Creative | Generate algorithmic art |
-
-### Invoking Skills
-
-Skills are invoked using the `InvokeSkill` tool:
-
-```
-InvokeSkill(skillId="docx", taskDescription="Create a document with title 'Report' and content 'Hello World'")
-```
-
-## Skill System Files
-
-Skills are defined in the `skills/` directory with the following structure:
-
-```
-skills/
-├── skills/
-│   ├── {skill-name}/
-│   │   ├── SKILL.md          # Skill definition
-│   │   ├── LICENSE.txt
-│   │   └── scripts/          # Implementation scripts
-│   └── ...
-├── spec/
-│   └── agent-skills-spec.md  # Skill specification
-└── template/
-    └── SKILL.md              # Skill template
-```
-
-## Best Practices
-
-1. **Security**: Review MCP server permissions before installing
-2. **Versioning**: Use specific versions for reproducibility
-3. **Testing**: Test workflows and skills in a safe environment first
-4. **Documentation**: Document custom SKILLs and workflows
-5. **Skill Selection**: Use InvokeSkill tool for specialized tasks
-
-## Related Documentation
-
-- [Architecture Overview](./overview.md)
-- [Tool System Design](./tool-system-design.md)
-- [CLI Commands](../cli/commands.md)
-- [Smart Mode Documentation](../smart-mode.md)
+# SKILL, Workflow and MCP Integration Guide
+
+This document explains how to integrate external skills, workflows, and MCP servers with xAgent CLI.
+
+## Overview
+
+xAgent CLI supports extensibility through:
+- **SKILL**: Reusable task modules
+- **Workflow**: Predefined automation sequences
+- **MCP (Model Context Protocol)**: External tool servers
+
+---
+
+## MCP Integration
+
+### What is MCP?
+
+MCP (Model Context Protocol) is an open standard that enables AI assistants to connect with external tools and data sources. xAgent uses MCP to integrate additional tools beyond its built-in set.
+
+### Configuring MCP Servers
+
+Add MCP servers to your settings file (`~/.xagent/settings.json`):
+
+```json
+{
+  "mcpServers": {
+    "filesystem": {
+      "command": "npx",
+      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/directory"],
+      "transport": "stdio"
+    },
+    "github": {
+      "command": "npx",
+      "args": ["-y", "@modelcontextprotocol/server-github"],
+      "transport": "stdio"
+    },
+    "custom-server": {
+      "url": "http://localhost:3000/mcp",
+      "transport": "http"
+    }
+  }
+}
+```
+
+### MCP Server Configuration
+
+```typescript
+interface MCPServerConfig {
+  command?: string;           // Command for stdio transport
+  args?: string[];           // Command arguments
+  env?: Record<string, string>;  // Environment variables
+  cwd?: string;              // Working directory
+  url?: string;              // URL for HTTP/SSE transport
+  transport?: 'stdio' | 'sse' | 'http';
+  authToken?: string;        // Bearer token for authentication
+  headers?: Record<string, string>;  // Custom headers
+  timeout?: number;          // Tool call timeout in ms (default: 30000)
+}
+```
+
+### MCP Tool Naming
+
+MCP tools are automatically registered with the ToolRegistry and available to agents. To distinguish MCP tools from built-in tools, MCP tool names use a prefix pattern:
+
+```
+{serverName}__{toolName}
+```
+
+For example:
+- `filesystem__read_file`
+- `github__create_issue`
+
+The prefix is derived from the server name in your configuration (the key in `mcpServers`).
+
+### Adding MCP Servers
+
+#### Interactive Mode (Recommended)
+
+```bash
+xagent start
+# Then use:
+/mcp add
+```
+
+#### CLI Command (Non-interactive)
+
+xAgent CLI supports adding MCP servers via command line without interactive prompts:
+
+```bash
+# Add with stdio transport
+xagent mcp --add <name> -t stdio -c <command> --args <args>
+
+# Add with HTTP transport
+xagent mcp --add <name> -t http -u <url> [-k <token>] [--header <key:value>]
+```
+
+**Options:**
+
+| Short | Long | Description |
+|-------|------|-------------|
+| `-a` | `--add [name]` | Server name (auto-generated if not provided) |
+| `-t` | `--transport <type>` | Transport type: `stdio` or `http` |
+| `-c` | `--command <cmd>` | Command for stdio transport |
+| | | `--args <args>` | Arguments for stdio transport (comma-separated) |
+| `-u` | `--url <url>` | URL for HTTP transport |
+| `-k` | `--token <token>` | Bearer authentication token |
+| | | `--header <key:value>` | Custom header (can be used multiple times) |
+| `-y` | `--yes` | Skip confirmation prompt |
+
+**Examples:**
+
+```bash
+# GitHub MCP server (stdio)
+xagent mcp --add github -t stdio -c "npx" --args "-y,@modelcontextprotocol/server-github"
+
+# Filesystem MCP server (stdio)
+xagent mcp --add filesystem -t stdio -c "npx" --args "-y,@modelcontextprotocol/server-filesystem,/path/to/dir"
+
+# Custom HTTP server with auth
+xagent mcp --add custom -t http -u "https://localhost:3000/mcp" -k "bearer-token"
+
+# Custom headers
+xagent mcp --add custom -t http -u "https://example.com/mcp" --header "X-Custom-Header:value"
+
+# Skip confirmation
+xagent mcp --add github -t stdio -c "npx" --args "-y,@modelcontextprotocol/server-github" -y
+```
+
+**Transport Types:**
+
+- **stdio**: Spawns the MCP server as a subprocess. Used for local servers like `@modelcontextprotocol/server-github`, `@modelcontextprotocol/server-filesystem`.
+- **http**: Streamable HTTP transport. Used for remote MCP servers.
+
+**Complete Interactive Flow Example:**
+
+```
+> /mcp add
+? Enter MCP server name: github
+? Select transport type: Stdio (stdin/stdout)
+? Enter command (for stdio transport): npx
+? Enter arguments (comma-separated, for stdio transport): -y, @modelcontextprotocol/server-github
+✅ MCP server 'github' added and connected successfully
+```
+
+**For HTTP/SSE Transport:**
+
+```
+> /mcp add
+? Enter MCP server name: custom-api
+? Select transport type: HTTP (POST)
+? Enter server URL (for HTTP/SSE/HTTP transport): https://api.example.com/mcp
+? Enter authentication token (optional):
+? Enter custom headers as JSON (optional): {"X-Custom-Header": "value"}
+✅ MCP server 'custom-api' added and connected successfully
+```
+
+#### Manual Configuration
+
+Edit `~/.xagent/settings.json`:
+
+```json
+{
+  "mcpServers": {
+    "github": {
+      "command": "npx",
+      "args": ["-y", "@modelcontextprotocol/server-github"],
+      "transport": "stdio"
+    }
+  }
+}
+```
+
+Restart your session after editing.
+
+### Managing MCP Servers
+
+```bash
+# List all configured MCP servers
+xagent mcp --list
+
+# Remove an MCP server
+xagent mcp --remove <name>
+```
+
+---
+
+## Workflow System
+
+### What are Workflows?
+
+Workflows are predefined sequences of actions that automate complex tasks. They can be installed from the Workflow Market.
+
+### Installing Workflows
+
+```bash
+xagent workflow --add <workflow-id>
+```
+
+### Workflow Structure
+
+```typescript
+interface WorkflowConfig {
+  id: string;
+  name: string;
+  description: string;
+  version: string;
+  author: string;
+  agents: AgentConfig[];
+  commands: Record<string, string>;
+  mcpServers: Record<string, MCPServerConfig>;
+  xagentMd: string;
+  files: Record<string, string>;
+}
+```
+
+### Managing Workflows
+
+```bash
+# List installed workflows
+xagent workflow --list
+
+# Remove a workflow
+xagent workflow --remove <workflow-id>
+```
+
+---
+
+## Skill Integration
+
+### What are Skills?
+
+Skills are reusable task modules that extend xAgent's capabilities. Each skill is defined by a `SKILL.md` file and can be invoked during conversations.
+
+### Installing Skills
+
+xAgent supports installing skills from various sources:
+
+```bash
+# Install from local directory
+xagent skill --add ./my-skill
+
+# Install from GitHub repository
+xagent skill --add owner/repo
+
+# Install from GitHub URL
+xagent skill --add https://github.com/owner/repo
+
+# Install from specific path in repo
+xagent skill --add https://github.com/owner/repo/tree/main/skills/my-skill
+
+# Install skill with @ syntax
+xagent skill --add owner/repo@skill-name
+
+# Install from direct SKILL.md URL
+xagent skill --add https://example.com/skill.md
+```
+
+### Managing Skills
+
+```bash
+# List all installed skills
+xagent skill --list
+
+# Remove a user-installed skill
+xagent skill --remove <skill-name>
+```
+
+**Note:** Built-in skills `find-skills` cannot be removed.
+
+### Invoking Skills
+
+Skills are invoked using the `InvokeSkill` tool:
+
+```
+InvokeSkill(skillId="docx", taskDescription="Create a document with title 'Report'")
+```
+
+### Configuring Skill Paths
+
+By default, user skills are stored in `~/.xagent/skills`. You can customize this path in your settings(`~/.xagent/settings.json`):
+
+```json
+{
+  "userSkillsPath": "/path/to/your/skills"
+}
+```
+
+---
+
+## Best Practices
+
+1. **Security**: Review MCP server permissions before installing
+2. **Versioning**: Use specific versions for reproducibility
+3. **Testing**: Test workflows and skills in a safe environment first
+4. **Documentation**: Document custom SKILLs and workflows
+
+---
+
+## Related Documentation
+
+- [Architecture Overview](./overview.md)
+- [Tool System Design](./tool-system-design.md)
+- [CLI Commands](../cli/commands.md)
+- [Smart Mode Documentation](../smart-mode.md)