The two approaches
Tools and MCP servers
Tools are discrete, callable actions: search a knowledge base, query a database, call an external API, run a code snippet. You attach them directly to Airia Agent, and it invokes them as needed during a conversation. MCP (Model Context Protocol) servers are a standardized way to expose multiple tools from a single external server. Connecting an MCP server makes all of its tools available to Airia Agent at once. Use tools or MCP servers when:- The capability is a single, well-defined action (fetch, query, write, execute)
- You want low-latency access to a frequently used integration
- The tool does not require its own persona, conversation context, or specialized reasoning
Sub-agents
Sub-agents are fully built Airia agents from your catalog that Airia Agent can invoke during a conversation. When Airia Agent determines a task is best handled by a specialized agent, it delegates the conversation turn to that agent and returns the result. Use sub-agents when:- You already have a purpose-built agent (e.g., an HR assistant, a code reviewer) that should remain independently accessible in the catalog
- The capability requires multi-step reasoning, its own system prompt, or specialized knowledge
- You want to route requests between distinct domains without exposing multiple entry points to end users
Decision guide
| What you want to do | Recommended approach |
|---|---|
| Call an external REST API | Tool |
| Query a database | Tool |
| Execute code in the sandbox | Tool |
| Use an external MCP server (e.g., a file system, browser, or custom server) | MCP server |
| Delegate to an existing HR bot | Sub-agent |
| Delegate to an existing code-review agent | Sub-agent |
| Reuse an agent end users also access directly from the catalog | Sub-agent |
| Add a capability that needs its own persona or specialized knowledge | Sub-agent |
Using both together
Tools and sub-agents are additive. Airia Agent has access to all configured tools and all configured sub-agents simultaneously. At runtime, the agent chooses the most appropriate capability based on the user’s request.If you are unsure which approach to use, start with a sub-agent. Tools can always be added later once a capability is built out, tested, and validated as a named agent. Reverting a tool-based approach is harder once users rely on it.
Ready to configure?
