Tool Integration Guide
Learn how to integrate external tools and services with your Agent5ive agents.
Important: Agent Tools Cannot Be Modified After Generation
Once an agent is generated, you cannot edit the selected tools. This limitation applies to all tool categories including:
- Standard Tools (Web Search, Maps, etc.)
- Image & File Upload options
- Voice Chat capabilities
- Advanced Tools (MCP)
To change tools or configuration, you must use the "Start New Chat with this Vision" button on the main Builder Chat to create a new agent with your desired tools.
Understanding Tool Integration
Tool integration allows your agents to perform actions or retrieve information beyond their base language model capabilities. Agent5ive uses Composio to integrate external tools with your agents, enabling them to interact with APIs, search the web, process data, and more.
How Tool Integration Works
- Tool Selection - During agent creation, you select tools that match your agent's purpose
- Configuration - The platform configures the agent to use these tools appropriately
- Runtime Integration - When interacting with users, the agent determines when to use tools
- Tool Execution - The agent calls the appropriate tool with the necessary parameters
- Result Processing - The agent processes the tool's output and incorporates it into its response
Tool Approval Policy System
The Tool Approval Policy System gives you fine-grained control over how tools are executed in your agent deployments. You can configure whether tools require user confirmation before execution, are automatically approved, or are disabled entirely.
Applies to SDK/IFRAME/REACT, Embedded Widgets, MCP, and Voice
How Tool Approval Works
When configuring your agent's tools, you can set approval policies at two levels: a default rule for all tools, and custom rules for individual tools. This gives you flexibility to balance automation with user control.
All tools execute immediately without user confirmation. Best for trusted environments and streamlined user experiences.
Users must confirm each tool execution. Provides maximum transparency and control over agent actions.
Set individual approval policies for each tool. Allows you to require approval for sensitive operations while auto-approving routine tasks.
Configuring Tool Approval Policies
Example showing custom approval rules for standard tools: Google Maps Search set to auto-approve, Image Search requiring approval, and Web Browsing disabled.
Configuration Steps:
- Select Tool Approval Mode - Choose from "Automatically approve every tool", "Require approval for every tool", or "Custom rules"
- Set Default Rule - When using custom rules, set a default behavior for tools without explicit overrides
- Configure Individual Tools - For each tool, choose:
- Auto (use without confirmation) - Tool executes immediately
- Require approval - User must confirm before tool executes
- Disable - Tool is not available for this deployment
Example showing custom approval rules for MCP tools: TODOIST_CREATE_TASK and TODOIST_GET_ALL_COMMENTS require approval, TODOIST_DELETE_TASK is disabled, while TODOIST_GET_ALL_LABELS and TODOIST_GET_ALL_PROJECTS auto-approve.
Best Practice for MCP Tools:
Consider requiring approval for destructive operations (like DELETE or UPDATE) while auto-approving read-only operations (like GET or SEARCH). This balances user control with convenience.
Use Cases for Tool Approval Policies
- Read-only operations - Web searches, data retrieval, information lookups
- Trusted environments - Internal company tools with limited risk
- Streamlined UX - Consumer-facing agents where speed matters
- Low-risk tools - Maps searches, weather lookups, general information
- Destructive operations - Deleting data, modifying records, sending emails
- Financial transactions - Payment processing, fund transfers, purchases
- Sensitive data access - Personal information, confidential documents
- External communications - Sending messages, posting content, API calls
Important Considerations
- Tool approval policies apply to SDK/IFRAME/REACT, embedded widgets, MCP integrations, and voice conversations
- WhatsApp and Telegram integrations always auto-approve tools
- Disabled tools are completely unavailable to the agent in that deployment
- Default rules apply to any tools without explicit custom rules
- Consider the user experience when requiring approval for frequently-used tools
Available Tools
Standard Tools Now Available in Voice
Agent5ive provides a comprehensive set of search tools that allow your agents to find and retrieve information from various sources. These tools enhance your agent's ability to provide accurate and up-to-date information in both text and voice conversations.
Available Search Tools:
- Web Browsing - Search the web for information
Enables your agent to find general information across the internet.
- Google Maps Search - Search for locations on Google Maps
Helps your agent find specific locations, businesses, and points of interest.
- Google Maps Directions - Get directions and location information
Allows your agent to provide navigation guidance between locations.
- Image Search - Search for images on the web
Enables your agent to find relevant images for visual references.
- News Search - Search for news articles
Helps your agent stay current with the latest news and developments.
- Finance Search - Search for financial information
Provides access to financial data, stock information, and market trends.
- Scholar Search - Search for scholarly articles
Enables your agent to access academic and research publications.
- Trends Search - Search for trends and insights
Helps your agent identify popular topics and trending subjects.
- Shopping Search - Search for shopping deals
Allows your agent to find products, compare prices, and identify deals.
- Event Search - Search for events and activities
Enables your agent to find upcoming events, concerts, conferences, and activities.
- Amazon Search - Search for products on Amazon
Find and compare products available on Amazon marketplace.
- DuckDuckGo Search - Search the web with DuckDuckGo
Privacy-focused web search for general information retrieval.
- Exa Answer - Get answers with citations using the Exa API
Provides well-researched answers with proper source citations.
- Exa Similar Link - Find similar links with the Exa API
Discover related content and resources based on similarity.
- Flights Search - Search for flights
Find and compare flight options for travel planning.
- Hotels Search - Search for hotels and vacation rentals
Locate accommodation options for travel destinations.
- Tavily Search - Perform a search with the Tavily API
Advanced search capabilities using Tavily's specialized search engine.
- TripAdvisor Search - Search for travel recommendations on TripAdvisor
Find travel recommendations, reviews, and destination information.
- Walmart Search - Search for products on Walmart
Find and compare products available at Walmart.
- Mail Service - Send emails on behalf of the agent
Enables your agent to send emails to users or other recipients. Best practice: Include email template instructions in your system prompt or knowledge base notes to ensure consistent and appropriate email formatting.
Best Practice for Mail Service:
Define email templates and formatting guidelines in your agent's system prompt or knowledge base. For example: "When sending confirmation emails, use a professional tone and include: greeting, confirmation details, next steps, and signature." This ensures your agent sends consistent, well-formatted emails that align with your brand voice.
Example Usage:
A user asks your travel assistant about attractions in Paris. The agent uses the Google Maps Search tool to find popular tourist destinations, then uses the Web Browsing tool to gather information about opening hours and ticket prices.
Use With Caution
Agent Persistent Memory enables your agents to remember information across different conversations and threads. This powerful feature is available only for Builder, Pro, and Business tier subscribers, allowing your agents to build a knowledge base over time.
How Agent Persistent Memory Works:
- Write Operation - Store information under specific keys
Requires 'key' and 'value' parameters. Related data is concatenated under a single key.
- Read Operation - Retrieve information stored under a specific key
Requires 'key' parameter to fetch the stored information.
- Read All Operation - Retrieve all stored information (up to 100 entries)
Returns up to 100 entries with no parameters needed.
- Search Operation - Find information based on a query
Searches both keys and values using the 'query' parameter. Optionally takes 'limit' parameter (default: 10, max: 50).
- Memory Scoping - Control the scope of stored information
All operations can optionally take 'threadId' to scope memory to a specific conversation; otherwise, memory is global to the agent deployment.
Example Usage:
A customer support agent can remember user preferences across multiple conversations. When a user mentions they prefer email communication, the agent stores this preference and applies it in future interactions without asking again.
Best Practices:
- Use descriptive key names that clearly indicate the type of information stored
- Structure your data consistently to make retrieval and searching more effective
- Include validation in your system prompt to prevent storing sensitive or inappropriate information
- Consider using thread-specific memory for personal preferences and global memory for general knowledge
- Regularly review stored information to ensure accuracy and relevance
Important Considerations:
- Privacy Impact - Information stored in agent memory persists across conversations and may be accessed in future interactions
- System Prompt Guidance - Your system prompt should clearly define what information the agent should store and how it should use that information
- User Consent - Consider informing users that the agent may remember information from previous conversations
- Data Accuracy - Stored information may become outdated; consider implementing a verification mechanism
- Memory Limitations - There are limits to how much information can be stored and retrieved
Configuration Note:
Image Upload & Vision must be enabled during initial agent creation. Once the agent is generated, this setting cannot be modified. To change this setting, you'll need to create a new agent using "Start New Chat with this Vision".
Agent5ive allows you to enable image upload and vision capabilities for your agents. This powerful feature lets users share images with your agent, which can then analyze and respond to the visual content.
How Image Vision Works:
- Enable the Feature - Toggle on "Enable Image Upload & Vision" during creation or in the Edit Details section
This requires a vision-capable model (like Gemini Flash).
- User Uploads - Users can upload images directly in the chat interface
The upload button appears in the chat input area when this feature is enabled.
- Image Analysis - Your agent can recognize content in images
For example, a travel agent can identify landmarks or destinations from photos.
- Contextual Responses - Your agent responds based on the image content
The agent can provide information, recommendations, or ask follow-up questions related to what it sees.
Example Usage:
A user shares a photo of the Eiffel Tower with your travel agent. The agent recognizes Paris, and asks specific questions about when they're planning to visit, their budget, and interests to help create a personalized Paris itinerary.
Agent5ive allows you to enable document file upload capabilities for your agents. This feature lets users share documents with your agent, which can then analyze and respond to the content within those files.
Configuration Note:
File Upload capabilities must be enabled during initial agent creation. Once the agent is generated, this setting cannot be modified. To change this setting, you'll need to create a new agent using "Start New Chat with this Vision".
How File Upload Works:
- Enable the Feature - Toggle on "Enable File Upload (PDF, DOCX, TXT)" during creation or in the Edit Details section
This allows your agent to process various document formats.
- User Uploads - Users can upload document files directly in the chat interface
The upload button appears in the chat input area when this feature is enabled.
- Document Processing - Your agent can extract and analyze content from uploaded files
For example, a research assistant can analyze data from uploaded PDFs or documents.
- Contextual Responses - Your agent responds based on the document content
The agent can provide summaries, answer questions, or perform tasks based on the information in the uploaded files.
Example Usage:
A user shares a research paper PDF with your academic assistant agent. The agent analyzes the paper, extracts key findings, and can answer specific questions about the methodology or conclusions presented in the document.
Experimental Feature
Configuration Note:
MCP integration must be configured during initial agent creation. Once the agent is generated, this setting cannot be modified. To change this setting, you'll need to create a new agent using "Start New Chat with this Vision".
The Advanced Tools (MCP) feature allows you to connect your agents to external Server-Sent Events (SSE/MCP [Http Streamable]) hosted servers, such as those provided by Composio, using the Model Context Protocol (MCP). This enables your agents to access a wide range of third-party services and APIs that aren't available through standard tools.
Available Tutorials:
Follow these step-by-step tutorials to set up specific MCP integrations with your agents.
Getting Started with MCP Servers:
There are two main platforms for accessing MCP servers:
Composio MCP Dashboard
Official platform for enterprise-grade MCP integrations with popular services like Gmail, Google Calendar, and GitHub.
- Visit the Composio MCP Dashboard
- Create an account or sign in
- Browse available integrations
- Create a server for your desired integration
- Configure authentication and permissions
- Copy the MCP Server URL
Smithery.ai
Community platform with hundreds of MCP servers including Exa search, Context7 documentation, Browserbase automation, and more.
Important: Use "Get URL with keys instead"
- Visit Smithery.ai
- Browse or search for MCP servers
- Select a server that matches your needs
- Click "Get URL with keys instead" (required for most servers)
- Copy the generated URL with embedded keys
- Paste directly into Agent5ive MCP configuration
See our Smithery.ai tutorial for detailed instructions.
For detailed documentation on creating and configuring MCP servers, visit the tutorials page.
Example Usage:
You create a personal assistant agent and connect it to a Google Calendar MCP server from Composio. Users can now ask your agent to "Schedule a meeting with John on Friday at 2 PM" or "What meetings do I have tomorrow?", and the agent will use the Google Calendar API to create events or retrieve schedule information.
Important Considerations:
- MCP servers require proper authentication and authorization to access third-party services
- Users may need to grant permissions for your agent to access their data through the connected services
- Some integrations may have rate limits or usage restrictions imposed by the service provider
- As an experimental feature, the MCP integration may change or be updated in future releases
- Always test thoroughly before deploying agents with MCP tools in production environments
Voice Chat Now Stable
Voice Chat Now Supports Standard Tools
Configuration Note:
Voice capabilities must be enabled during initial agent creation. Once the agent is generated, this setting cannot be modified. To change this setting, you'll need to create a new agent using "Start New Chat with this Vision".
The Real-time Voice Chat feature allows your agents to listen and respond using natural speech, creating a more intuitive and accessible interaction experience. This feature leverages OpenAI's advanced GPT-realtime model to provide high-quality voice interactions with support for MCP tools and agent communication.
How Voice Chat Works:
- Enable the Feature - Expand the "Real-time Voice Chat" section and toggle it on during creation or in the Edit Details section
- Select a Voice - Choose from available voice options to personalize your agent's speaking style
- Provide API Key - Enter your OpenAI API key to enable the voice processing capabilities
- Voice Interaction - Users can speak to your agent using their microphone and hear responses in the selected voice
- MCP Tool Support - Voice conversations can use experimental MCP tools for enhanced functionality
- Agent Communication - Voice agents can communicate with other agents as tools
Example Usage:
A user can have a hands-free conversation with your customer support agent while multitasking. They can ask questions verbally and receive spoken responses, making the interaction more natural and accessible.
OpenAI API Key Required
Important Note:
Once an agent is generated, you cannot modify which agents are used as tools. If you need to change agent tools, you must use the "Start New Chat with this Vision" button to create a new agent with your desired configuration.
Agent5ive now allows you to use your publicly deployed agents as tools for other agents. This powerful feature enables you to create specialized agents that can work together, with one agent calling on the expertise of others when needed.
How Multi-Agent Integration Works:
- Deploy and Publish Specialized Agents - First, create, deploy, and publish agents with specific expertise
Agents must be published publicly to be available as tools for other agents.
- Select Agents as Tools - When creating a new agent, you'll see a "Select Your Agents as Tools" section
This section allows you to select from your publicly deployed agents to use as tools.
- Multi-Agent Collaboration - Your primary agent can now call on these specialized agents when needed
For example, a fitness blog writer agent might consult a nutrition expert agent for diet recommendations.
- Seamless Integration - The process is transparent to end users, who simply interact with the primary agent
Behind the scenes, the primary agent is delegating specific questions to specialized agents.
Example Usage:
A user asks your fitness coach agent for a complete wellness plan. The fitness coach agent uses a nutrition specialist agent to create the diet portion, a workout planner agent to design the exercise routine, and a meditation guide agent to suggest mindfulness practices, then combines all this information into a comprehensive response.
Important Notes:
- Only publicly published agents can be used as tools for other agents
- Agents with the "Multi Agent" badge are using at least one other agent as a tool
- Multi-agent functionality is currently not available in voice conversations, but will work in text conversations (even with voice-enabled agents)
- When using a voice-enabled agent that also has agent tools, the text input box will still utilize the agent tools while voice interactions will not
Custom tool integration via Composio is planned for a future update of Agent5ive. This feature will allow you to create your own tools that connect to your specific systems, APIs, or data sources.
Important Note:
Custom tool integration is not currently available. The platform currently supports only the pre-built tools listed above. We'll update this documentation when custom tool integration becomes available.
Current Limitations:
- No Local File Access - Agents cannot access files on your computer or local devices
- No Direct Database Connections - Agents cannot directly connect to your databases
- No Custom API Integration - Custom API connections are not currently supported
If you need your agent to have access to specific information, consider using the Knowledge Base feature to upload documents or provide URLs that contain the necessary information.