Expand your agent's capabilities
Expand your agent's capabilities
15 min
Gaëlle CaplierRemember how you can add extra capabilities to your agents on the fly, directly during your conversations? Remember how those extra capabilities don't persist in new conversations with the same agent? How about we actually add those capabilities directly in the agent to make them permanent!
1. Agent capabilities
You can add capabilities to your agent from the section "Capabilities and Knowledge" in the agent builder. There are two types of capabilities: tools and skills.
- Tools are extra capabilities you can give to your agent, including web browsing, creating images, connecting to your work applications, and more.
- Skills are small packages of tools and instructions that can be shared across agents.
For the remainder of this course, we will use both tools and skills interchangeably.
2. MCP tools - to connect to your work applications
2.1. Model Context Protocol - what are we talking about?
The Model Context Protocol (MCP) is a standardized way for AI systems to connect with the tools and software your organization already uses.
Think of it like a universal adapter: before MCP, every time you wanted your AI to interact with a different system (your customer database, project management tool, internal documentation, etc), someone had to build a custom connection from scratch. MCP changes this by providing a common language that both the AI and your systems can understand.
Once a tool is set up to work with MCP, any AI platform that supports the protocol can use it immediately, without additional custom work. This makes it much simpler and faster to build AI agents that can actually take action across your business systems, rather than just answering questions.
2.2. MCP tools in Dust
Because MCP is a universal adapter, we can connect Dust to pretty much any application - on the condition that they expose a MCP server (e.g. a way to connect through the protocol). This means that:
- We have a lot of MCP tools available out of the box, and we add new ones regularly
- You can also connect your own custom MCP servers to expand Dust capabilities beyond what we offer by default!
⚠️ Only workspace administrators can add and edit MCP tools. If you're not an admin and you're wondering which tools are available on your workspace, you can always ask@dust what tools do you have access to?
Here's a (non-exhaustive, because we're always adding more!) list of MCP tools available in Dust:
Tool name | What the agent will do |
|---|---|
GMail | Search email, draft email, send email |
Outlook | Search email, draft email |
Google Calendar & Outlook Calendar | Search, update and create events in Google Calendar or Outlook Calendar |
Notion | Search info, create and update pages in Notion |
Slack | Search info, post messages in Slack |
Teams | Search info, post messages in Teams |
GitHub | Search, update, comment, and create pull requests and issues in GitHub. |
Confluence | Search, create, and update pages in Confluence. |
HubSpot | Search, update and create info in Hubspot. |
Zendesk | Search and retrieve support tickets, view ticket metrics and conversations, and draft replies. |
Salesforce | Search, update and create info in Salesforce. |
2.3. Other capabilities for your agents
Beyond MCP tools, you can also expand your agents' capabilities in other ways.
2.3.1. Web Search & Browse
Enables agents to search Google and browse webpages for up-to-date information beyond your internal data sources. The agent can access open websites (not behind paywalls or authentication).
Use cases: Market research, competitive intelligence, finding latest news about customers, gathering public information.
2.3.2. Frames
Transforms static text responses into interactive visualizations. Agents can create reports with live data, dashboards, calculators, data visualizations, or any interactive element you describe.
Frames are shareable files with secure links for internal or external sharing.
Use cases: Executive reports, client deliverables, team dashboards, interactive data exploration.
2.3.3. Image Generation & Editing
Generates images from text prompts or edits existing images using natural language instructions.
Use cases: Marketing visuals, concept art, product mockups, diagrams, photo editing.
2.3.4. Go Deep
Hands off complex research questions to @deep-dive for comprehensive analysis. When your agent encounters a question requiring extensive investigation, it can delegate to @deep-dive while preserving context.
Use cases: Complex research across multiple data sources, deep analysis requiring SQL queries and web research, market intelligence, competitive analysis.
2.3.5. Agent Memory
Enables agents to remember information about a specific user across conversations, providing personalized experiences. The agent automatically decides what to remember, or users can explicitly ask it to remember specific details.
💡 Memories are private and user-scoped. Each user has their own memory space with each agent. Your memories are only visible to you.
Use cases: Personal preferences, work context, recurring tasks, improving agent responses over time.
2.3.6. Run Agent
Allows agents to call other specialized agents as tools. Two modes: background execution (agent runs separately and returns output) or conversation handoff (sub-agent takes over and responds directly).
Use cases: Creating modular agent architectures, delegating subtasks, routing users to specialized agents, preventing context exhaustion in complex workflows.
2.3.7. Create File
Generates files in various formats (PDF, Word, Excel, PowerPoint, CSV, Markdown, HTML, XML, images) and converts between formats.
Use cases: Creating reports, converting documents, generating spreadsheets, exporting data.
2.3.8. Speech generator
Converts text into natural-sounding audio with configurable voices, languages, and speaking styles. Supports single-voice narration or multi-speaker dialogues.
Use cases: Podcast narration, audio versions of written content, training materials, multilingual voice content.
2.3.9. Sound studio
Generates sound effects and non-verbal audio elements to enhance audio content.
Use cases: Adding sound design to podcasts, creating ambient audio for training materials, generating custom sound effects.
3. In action: Create an interactive newsletter agent
This use case leverages the following capabilities:
Create Frames, Create Images, Slack MCP tool. It also covers how to use Frame templates.Conclusion
Thank you for following this course with us! You now know the best practices for creating your own custom agents - from writing good custom instructions, to processing your data the right way, and to selecting extra tools and capabilities. See you again soon!
Test Your Knowledge
Answer 5 questions — score 3 or more to pass
Ready to test your understanding of this chapter?