Files
n8n-mcp/dist/mcp/tool-docs/guides/ai-agents-guide.js
2025-12-05 12:01:31 +01:00

739 lines
20 KiB
JavaScript

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.aiAgentsGuide = void 0;
exports.aiAgentsGuide = {
name: 'ai_agents_guide',
category: 'guides',
essentials: {
description: 'Comprehensive guide to building AI Agent workflows in n8n. Covers architecture, connections, tools, validation, and best practices for production AI systems.',
keyParameters: [],
example: 'Use tools_documentation({topic: "ai_agents_guide"}) to access this guide',
performance: 'N/A - Documentation only',
tips: [
'Start with Chat Trigger → AI Agent → Language Model pattern',
'Always connect language model BEFORE enabling AI Agent',
'Use proper toolDescription for all AI tools (15+ characters)',
'Validate workflows with n8n_validate_workflow before deployment',
'Use includeExamples=true when searching for AI nodes',
'Check FINAL_AI_VALIDATION_SPEC.md for detailed requirements'
]
},
full: {
description: `# Complete Guide to AI Agents in n8n
This comprehensive guide covers everything you need to build production-ready AI Agent workflows in n8n.
## Table of Contents
1. [AI Agent Architecture](#architecture)
2. [Essential Connection Types](#connections)
3. [Building Your First AI Agent](#first-agent)
4. [AI Tools Deep Dive](#tools)
5. [Advanced Patterns](#advanced)
6. [Validation & Best Practices](#validation)
7. [Troubleshooting](#troubleshooting)
---
## 1. AI Agent Architecture {#architecture}
### Core Components
An n8n AI Agent workflow typically consists of:
1. **Chat Trigger**: Entry point for user interactions
- Webhook-based or manual trigger
- Supports streaming responses (responseMode)
- Passes user message to AI Agent
2. **AI Agent**: The orchestrator
- Manages conversation flow
- Decides when to use tools
- Iterates until task is complete
- Supports fallback models for reliability
3. **Language Model**: The AI brain
- OpenAI GPT-4, Claude, Gemini, etc.
- Connected via ai_languageModel port
- Can have primary + fallback for reliability
4. **Tools**: AI Agent's capabilities
- HTTP Request, Code, Vector Store, etc.
- Connected via ai_tool port
- Each tool needs clear toolDescription
5. **Optional Components**:
- Memory (conversation history)
- Output Parser (structured responses)
- Vector Store (knowledge retrieval)
### Connection Flow
**CRITICAL**: AI connections flow TO the consumer (reversed from standard n8n):
\`\`\`
Standard n8n: [Source] --main--> [Target]
AI pattern: [Language Model] --ai_languageModel--> [AI Agent]
[HTTP Tool] --ai_tool--> [AI Agent]
\`\`\`
This is why you use \`sourceOutput: "ai_languageModel"\` when connecting components.
---
## 2. Essential Connection Types {#connections}
### The 8 AI Connection Types
1. **ai_languageModel**
- FROM: OpenAI Chat Model, Anthropic, Google Gemini, etc.
- TO: AI Agent, Basic LLM Chain
- REQUIRED: Every AI Agent needs 1-2 language models
- Example: \`{type: "addConnection", source: "OpenAI", target: "AI Agent", sourceOutput: "ai_languageModel"}\`
2. **ai_tool**
- FROM: Any tool node (HTTP Request Tool, Code Tool, etc.)
- TO: AI Agent
- REQUIRED: At least 1 tool recommended
- Example: \`{type: "addConnection", source: "HTTP Request Tool", target: "AI Agent", sourceOutput: "ai_tool"}\`
3. **ai_memory**
- FROM: Window Buffer Memory, Conversation Summary, etc.
- TO: AI Agent
- OPTIONAL: 0-1 memory system
- Enables conversation history tracking
4. **ai_outputParser**
- FROM: Structured Output Parser, JSON Parser, etc.
- TO: AI Agent
- OPTIONAL: For structured responses
- Must set hasOutputParser=true on AI Agent
5. **ai_embedding**
- FROM: Embeddings OpenAI, Embeddings Google, etc.
- TO: Vector Store (Pinecone, In-Memory, etc.)
- REQUIRED: For vector-based retrieval
6. **ai_vectorStore**
- FROM: Vector Store node
- TO: Vector Store Tool
- REQUIRED: For retrieval-augmented generation (RAG)
7. **ai_document**
- FROM: Document Loader, Default Data Loader
- TO: Vector Store
- REQUIRED: Provides data for vector storage
8. **ai_textSplitter**
- FROM: Text Splitter nodes
- TO: Document processing chains
- OPTIONAL: Chunk large documents
### Connection Examples
\`\`\`typescript
// Basic AI Agent setup
n8n_update_partial_workflow({
id: "workflow_id",
operations: [
// Connect language model (REQUIRED)
{
type: "addConnection",
source: "OpenAI Chat Model",
target: "AI Agent",
sourceOutput: "ai_languageModel"
},
// Connect tools
{
type: "addConnection",
source: "HTTP Request Tool",
target: "AI Agent",
sourceOutput: "ai_tool"
},
{
type: "addConnection",
source: "Code Tool",
target: "AI Agent",
sourceOutput: "ai_tool"
},
// Add memory (optional)
{
type: "addConnection",
source: "Window Buffer Memory",
target: "AI Agent",
sourceOutput: "ai_memory"
}
]
})
\`\`\`
---
## 3. Building Your First AI Agent {#first-agent}
### Step-by-Step Tutorial
#### Step 1: Create Chat Trigger
Use \`n8n_create_workflow\` or manually create a workflow with:
\`\`\`typescript
{
name: "My First AI Agent",
nodes: [
{
id: "chat_trigger",
name: "Chat Trigger",
type: "@n8n/n8n-nodes-langchain.chatTrigger",
position: [100, 100],
parameters: {
options: {
responseMode: "lastNode" // or "streaming" for real-time
}
}
}
],
connections: {}
}
\`\`\`
#### Step 2: Add Language Model
\`\`\`typescript
n8n_update_partial_workflow({
id: "workflow_id",
operations: [
{
type: "addNode",
node: {
name: "OpenAI Chat Model",
type: "@n8n/n8n-nodes-langchain.lmChatOpenAi",
position: [300, 50],
parameters: {
model: "gpt-4",
temperature: 0.7
}
}
}
]
})
\`\`\`
#### Step 3: Add AI Agent
\`\`\`typescript
n8n_update_partial_workflow({
id: "workflow_id",
operations: [
{
type: "addNode",
node: {
name: "AI Agent",
type: "@n8n/n8n-nodes-langchain.agent",
position: [300, 150],
parameters: {
promptType: "auto",
systemMessage: "You are a helpful assistant. Be concise and accurate."
}
}
}
]
})
\`\`\`
#### Step 4: Connect Components
\`\`\`typescript
n8n_update_partial_workflow({
id: "workflow_id",
operations: [
// Chat Trigger → AI Agent (main connection)
{
type: "addConnection",
source: "Chat Trigger",
target: "AI Agent"
},
// Language Model → AI Agent (AI connection)
{
type: "addConnection",
source: "OpenAI Chat Model",
target: "AI Agent",
sourceOutput: "ai_languageModel"
}
]
})
\`\`\`
#### Step 5: Validate
\`\`\`typescript
n8n_validate_workflow({id: "workflow_id"})
\`\`\`
---
## 4. AI Tools Deep Dive {#tools}
### Tool Types and When to Use Them
#### 1. HTTP Request Tool
**Use when**: AI needs to call external APIs
**Critical Requirements**:
- \`toolDescription\`: Clear, 15+ character description
- \`url\`: API endpoint (can include placeholders)
- \`placeholderDefinitions\`: Define all {placeholders}
- Proper authentication if needed
**Example**:
\`\`\`typescript
{
type: "addNode",
node: {
name: "GitHub Issues Tool",
type: "@n8n/n8n-nodes-langchain.toolHttpRequest",
position: [500, 100],
parameters: {
method: "POST",
url: "https://api.github.com/repos/{owner}/{repo}/issues",
toolDescription: "Create GitHub issues. Requires owner (username), repo (repository name), title, and body.",
placeholderDefinitions: {
values: [
{name: "owner", description: "Repository owner username"},
{name: "repo", description: "Repository name"},
{name: "title", description: "Issue title"},
{name: "body", description: "Issue description"}
]
},
sendBody: true,
jsonBody: "={{ { title: $json.title, body: $json.body } }}"
}
}
}
\`\`\`
#### 2. Code Tool
**Use when**: AI needs to run custom logic
**Critical Requirements**:
- \`name\`: Function name (alphanumeric + underscore)
- \`description\`: 10+ character explanation
- \`code\`: JavaScript or Python code
- \`inputSchema\`: Define expected inputs (recommended)
**Example**:
\`\`\`typescript
{
type: "addNode",
node: {
name: "Calculate Shipping",
type: "@n8n/n8n-nodes-langchain.toolCode",
position: [500, 200],
parameters: {
name: "calculate_shipping",
description: "Calculate shipping cost based on weight (kg) and distance (km)",
language: "javaScript",
code: "const cost = 5 + ($input.weight * 2) + ($input.distance * 0.1); return { cost };",
specifyInputSchema: true,
inputSchema: "{ \\"type\\": \\"object\\", \\"properties\\": { \\"weight\\": { \\"type\\": \\"number\\" }, \\"distance\\": { \\"type\\": \\"number\\" } } }"
}
}
}
\`\`\`
#### 3. Vector Store Tool
**Use when**: AI needs to search knowledge base
**Setup**: Requires Vector Store + Embeddings + Documents
**Example**:
\`\`\`typescript
// Step 1: Create Vector Store with embeddings and documents
n8n_update_partial_workflow({
operations: [
{type: "addConnection", source: "Embeddings OpenAI", target: "Pinecone", sourceOutput: "ai_embedding"},
{type: "addConnection", source: "Document Loader", target: "Pinecone", sourceOutput: "ai_document"}
]
})
// Step 2: Connect Vector Store to Vector Store Tool
n8n_update_partial_workflow({
operations: [
{type: "addConnection", source: "Pinecone", target: "Vector Store Tool", sourceOutput: "ai_vectorStore"}
]
})
// Step 3: Connect tool to AI Agent
n8n_update_partial_workflow({
operations: [
{type: "addConnection", source: "Vector Store Tool", target: "AI Agent", sourceOutput: "ai_tool"}
]
})
\`\`\`
#### 4. AI Agent Tool (Sub-Agents)
**Use when**: Need specialized expertise
**Example**: Research specialist sub-agent
\`\`\`typescript
{
type: "addNode",
node: {
name: "Research Specialist",
type: "@n8n/n8n-nodes-langchain.agentTool",
position: [500, 300],
parameters: {
name: "research_specialist",
description: "Expert researcher that searches multiple sources and synthesizes information. Use for detailed research tasks.",
systemMessage: "You are a research specialist. Search thoroughly, cite sources, and provide comprehensive analysis."
}
}
}
\`\`\`
#### 5. MCP Client Tool
**Use when**: Need to use Model Context Protocol servers
**Example**: Filesystem access
\`\`\`typescript
{
type: "addNode",
node: {
name: "Filesystem Tool",
type: "@n8n/n8n-nodes-langchain.mcpClientTool",
position: [500, 400],
parameters: {
description: "Access file system to read files, list directories, and search content",
mcpServer: {
transport: "stdio",
command: "npx",
args: ["-y", "@modelcontextprotocol/server-filesystem", "/allowed/path"]
},
tool: "read_file"
}
}
}
\`\`\`
---
## 5. Advanced Patterns {#advanced}
### Pattern 1: Streaming Responses
For real-time user experience:
\`\`\`typescript
// Set Chat Trigger to streaming mode
{
parameters: {
options: {
responseMode: "streaming"
}
}
}
// CRITICAL: AI Agent must NOT have main output connections in streaming mode
// Responses stream back through Chat Trigger automatically
\`\`\`
**Validation will fail if**:
- Chat Trigger has streaming but target is not AI Agent
- AI Agent in streaming mode has main output connections
### Pattern 2: Fallback Language Models
For production reliability with fallback language models:
\`\`\`typescript
n8n_update_partial_workflow({
operations: [
// Primary model
{
type: "addConnection",
source: "OpenAI GPT-4",
target: "AI Agent",
sourceOutput: "ai_languageModel",
targetIndex: 0
},
// Fallback model
{
type: "addConnection",
source: "Anthropic Claude",
target: "AI Agent",
sourceOutput: "ai_languageModel",
targetIndex: 1
}
]
})
// Enable fallback on AI Agent
{
type: "updateNode",
nodeName: "AI Agent",
updates: {
"parameters.needsFallback": true
}
}
\`\`\`
### Pattern 3: RAG (Retrieval-Augmented Generation)
Complete knowledge base setup:
\`\`\`typescript
// 1. Load documents
{type: "addConnection", source: "PDF Loader", target: "Text Splitter", sourceOutput: "ai_document"}
// 2. Split and embed
{type: "addConnection", source: "Text Splitter", target: "Vector Store"}
{type: "addConnection", source: "Embeddings", target: "Vector Store", sourceOutput: "ai_embedding"}
// 3. Create search tool
{type: "addConnection", source: "Vector Store", target: "Vector Store Tool", sourceOutput: "ai_vectorStore"}
// 4. Give tool to agent
{type: "addConnection", source: "Vector Store Tool", target: "AI Agent", sourceOutput: "ai_tool"}
\`\`\`
### Pattern 4: Multi-Agent Systems
Specialized sub-agents for complex tasks:
\`\`\`typescript
// Create sub-agents with specific expertise
[
{name: "research_agent", description: "Deep research specialist"},
{name: "data_analyst", description: "Data analysis expert"},
{name: "writer_agent", description: "Content writing specialist"}
].forEach(agent => {
// Add as AI Agent Tool to main coordinator agent
{
type: "addConnection",
source: agent.name,
target: "Coordinator Agent",
sourceOutput: "ai_tool"
}
})
\`\`\`
---
## 6. Validation & Best Practices {#validation}
### Always Validate Before Deployment
\`\`\`typescript
const result = n8n_validate_workflow({id: "workflow_id"})
if (!result.valid) {
console.log("Errors:", result.errors)
console.log("Warnings:", result.warnings)
console.log("Suggestions:", result.suggestions)
}
\`\`\`
### Common Validation Errors
1. **MISSING_LANGUAGE_MODEL**
- Problem: AI Agent has no ai_languageModel connection
- Fix: Connect a language model before creating AI Agent
2. **MISSING_TOOL_DESCRIPTION**
- Problem: HTTP Request Tool has no toolDescription
- Fix: Add clear description (15+ characters)
3. **STREAMING_WITH_MAIN_OUTPUT**
- Problem: AI Agent in streaming mode has outgoing main connections
- Fix: Remove main connections when using streaming
4. **FALLBACK_MISSING_SECOND_MODEL**
- Problem: needsFallback=true but only 1 language model
- Fix: Add second language model or disable needsFallback
### Best Practices Checklist
✅ **Before Creating AI Agent**:
- [ ] Language model is connected first
- [ ] At least one tool is prepared (or will be added)
- [ ] System message is thoughtful and specific
✅ **For Each Tool**:
- [ ] Has toolDescription/description (15+ characters)
- [ ] toolDescription explains WHEN to use the tool
- [ ] All required parameters are configured
- [ ] Credentials are set up if needed
✅ **For Production**:
- [ ] Workflow validated with n8n_validate_workflow
- [ ] Tested with real user queries
- [ ] Fallback model configured for reliability
- [ ] Error handling in place
- [ ] maxIterations set appropriately (default 10, max 50)
---
## 7. Troubleshooting {#troubleshooting}
### Problem: "AI Agent has no language model"
**Cause**: Connection created AFTER AI Agent or using wrong sourceOutput
**Solution**:
\`\`\`typescript
n8n_update_partial_workflow({
operations: [
{
type: "addConnection",
source: "OpenAI Chat Model",
target: "AI Agent",
sourceOutput: "ai_languageModel" // ← CRITICAL
}
]
})
\`\`\`
### Problem: "Tool has no description"
**Cause**: HTTP Request Tool or Code Tool missing toolDescription/description
**Solution**:
\`\`\`typescript
{
type: "updateNode",
nodeName: "HTTP Request Tool",
updates: {
"parameters.toolDescription": "Call weather API to get current conditions for a city"
}
}
\`\`\`
### Problem: "Streaming mode not working"
**Causes**:
1. Chat Trigger not set to streaming
2. AI Agent has main output connections
3. Target of Chat Trigger is not AI Agent
**Solution**:
\`\`\`typescript
// 1. Set Chat Trigger to streaming
{
type: "updateNode",
nodeName: "Chat Trigger",
updates: {
"parameters.options.responseMode": "streaming"
}
}
// 2. Remove AI Agent main outputs
{
type: "removeConnection",
source: "AI Agent",
target: "Any Output Node"
}
\`\`\`
### Problem: "Agent keeps looping"
**Cause**: Tool not returning proper response or agent stuck in reasoning loop
**Solutions**:
1. Set maxIterations lower: \`"parameters.maxIterations": 5\`
2. Improve tool descriptions to be more specific
3. Add system message guidance: "Use tools efficiently, don't repeat actions"
---
## Quick Reference
### Essential Tools
| Tool | Purpose | Key Parameters |
|------|---------|----------------|
| HTTP Request Tool | API calls | toolDescription, url, placeholders |
| Code Tool | Custom logic | name, description, code, inputSchema |
| Vector Store Tool | Knowledge search | description, topK |
| AI Agent Tool | Sub-agents | name, description, systemMessage |
| MCP Client Tool | MCP protocol | description, mcpServer, tool |
### Connection Quick Codes
\`\`\`typescript
// Language Model → AI Agent
sourceOutput: "ai_languageModel"
// Tool → AI Agent
sourceOutput: "ai_tool"
// Memory → AI Agent
sourceOutput: "ai_memory"
// Parser → AI Agent
sourceOutput: "ai_outputParser"
// Embeddings → Vector Store
sourceOutput: "ai_embedding"
// Vector Store → Vector Store Tool
sourceOutput: "ai_vectorStore"
\`\`\`
### Validation Command
\`\`\`typescript
n8n_validate_workflow({id: "workflow_id"})
\`\`\`
---
## Related Resources
- **FINAL_AI_VALIDATION_SPEC.md**: Complete validation rules
- **n8n_update_partial_workflow**: Workflow modification tool
- **search_nodes({query: "AI", includeExamples: true})**: Find AI nodes with examples
- **get_node({nodeType: "...", detail: "standard", includeExamples: true})**: Node details with examples
---
*This guide is part of the n8n-mcp documentation system. For questions or issues, refer to the validation spec or use tools_documentation() for specific topics.*`,
parameters: {},
returns: 'Complete AI Agents guide with architecture, patterns, validation, and troubleshooting',
examples: [
'tools_documentation({topic: "ai_agents_guide"}) - Full guide',
'tools_documentation({topic: "ai_agents_guide", depth: "essentials"}) - Quick reference',
'When user asks about AI Agents, Chat Trigger, or building AI workflows → Point to this guide'
],
useCases: [
'Learning AI Agent architecture in n8n',
'Understanding AI connection types and patterns',
'Building first AI Agent workflow step-by-step',
'Implementing advanced patterns (streaming, fallback, RAG, multi-agent)',
'Troubleshooting AI workflow issues',
'Validating AI workflows before deployment',
'Quick reference for connection types and tools'
],
performance: 'N/A - Static documentation',
bestPractices: [
'Reference this guide when users ask about AI Agents',
'Point to specific sections based on user needs',
'Combine with search_nodes(includeExamples=true) for working examples',
'Validate workflows after following guide instructions',
'Use FINAL_AI_VALIDATION_SPEC.md for detailed requirements'
],
pitfalls: [
'This is a guide, not an executable tool',
'Always validate workflows after making changes',
'AI connections require sourceOutput parameter',
'Streaming mode has specific constraints',
'Fallback models require AI Agent node with fallback support'
],
relatedTools: [
'n8n_create_workflow',
'n8n_update_partial_workflow',
'n8n_validate_workflow',
'search_nodes',
'get_node'
]
}
};
//# sourceMappingURL=ai-agents-guide.js.map