In recent months, we’ve seen a growing number of teams use StackOne’s Unified APIs to build AI agents - think HR teammates, virtual recruiters, personal assistants, and beyond. This surge in agentic applications has been exciting, but it also exposed a key challenge: integrating SaaS API calls into agents is still harder than it should be.

Traditional SDKs and APIs weren’t designed for AI agents, forcing engineers to either hard-code API calls into tools, hack together custom OpenAPI-to-tool converters or rely on nascent MCP servers.

We realized there was an opportunity to provide a better developer experience. That’s why we released the StackOne ToolSet – an AI-native interface that lets AI agents seamlessly interact with hundreds of SaaS.

Before we dive into the ToolSet, let’s briefly look at why StackOne’s unified API provides a strong foundation for agents.

Agents and Unified API

StackOne’s platform has always focused on simplifying integrations by offering Unified APIs across over 200 enterprise applications. We normalize more than 16,000 fields into common, well defined schemas. Whether your agent is pulling data from Workday or Hibob, Salesforce or Attio, it returns clean, structured data with consistent fields and formats.

As the adage goes: garbage in, garbage out and it's no different when it comes to tool calling APIs. Disparate endpoint naming conventions, poorly named schema properties and a lack of descriptive usage guidance make for a subpar tool calling performance.

Unified inputs and outputs remove much of the complexity of dealing with SaaS tools and gives LLM-based agents the predictable interface to work with.

Data Efficiency: Another happy side effect of Unified APIs for agents is dramatically improved data efficiency.

For example an employee  in StackOne has attributes most apps would expect and few most wouldn’t. This makes the output of a Unified API many times more data efficient than the underlying API.

‍

‍

‍

‍The difference is stark: fetching a single employee record from Workday via StackOne uses less than 4% of the tokens required by Workday’s native SOAP API. ie., our responses are over 25× more concise.

And that’s just for one straightforward employee request, raw API outputs can easily becoming much larger for more complex queries or lists, quickly overwhelming an LLM’s context window. Since most LLMs, even those with extended context windows, experience reasoning degradation from around 10k tokens, many enterprise APIs are practically unusable by AI agents without this kind of preprocessing.

Simplified Tool Authentication: our Unified API abstracts all authentication management behind a single API key that agents can leverage to act on any SaaS connected by their customers (no matter the underlying authentication method of the SaaS). It turns out that this also neatly solves a pain point for authenticating SaaS tools for AI agents.

Real-Time Responses: StackOne was built on a real-time architecture. It acts as a pass-through proxy, transforming requests to and from the target systems in real time. When an agent creates or updates data in a system-of-record, it can immediately query those new records moments later. For AI agents, real-time capability is crucial; users expect an agent to "do something" and then immediately know it’s done, not have to wait for some background syncing job to complete.

Bridging the Gap from SDKs to Tools

Our standard SDKs are ergonomic and make it easy for developers to call unified API endpoints from their code. However, AI agents take action using a function calling or tool calling syntax, requiring some specification of tool inputs and a function to execute the tool. Converting numerous API calls or SDK methods into a LLM tools seemed like unnecessary overhead for our customers.

We looked at some emerging standards aiming to solving this interface problem:

• Wildcard AI’s agents.json format lets API providers describe endpoints with machine-readable instructions (similar to an OpenAPI spec)
• Model Context Protocol (MCP) from Anthropic defines a protocol for LLMs to interact with external tools and data sources.
• Open AI's SDK’s tool definition became an implicit standard thanks to its SDK becoming defacto for open source inference server libraries like vLLM.

The value of unification in StackOne is that our customers don't have to care about the underlying standards or their differences. They should be free to use our unified API through the interface of their choosing and from the library or framework which fits their use-case. We wanted a flexible solution that Agent developers could use today, without waiting for market adoption of an open source standard.

By taking a toolset-first approach, inspired by other API-first companies (such as Stripe and Vercel), we’re able to offer developers an improved experience that is immediately usable by any B2B agent. Importantly, this approach doesn’t lock us out of future standards.  We can always expose our toolset in whatever standard emerges and we’ll release MCP Servers soon.

Introducing the StackOne ToolSet

StackOne ToolSet is our answer to the agent integration challenge. It’s an open-source, experimental SDK tailor-made for AI agent development, available today in both TypeScript (Node.js) and Python. In short, StackOne ToolSet transforms StackOne’s unified API into a collection of AI tools that plug in smoothly with LLM frameworks.

While some early players in the AI integrations space have rushed to release catalogs of low-level tools for common apps, we’ve been quietly focused on building the best option for agents that actually want to interact reliably with SaaS tools of their customers.

The StackOne ToolSet is the result but also the very beginning of that work. It offers a significantly improved developer experience for anyone creating AI agents that need to take action on real, complex SaaS.

Key Features of StackOne ToolSet:

• Tool Filtering: You can easily filter which tools (endpoints) are exposed to your agent using simple glob patterns. This gives you fine-grained control over what an AI agent is allowed to do, focusing it on the actions it needs.
• Out-of-the-Box AI Library Integrations: The ToolSet works seamlessly with popular AI frameworks and SDKs for LLMs, including OpenAI’s function calling, LangChain, CrewAI, and Vercel’s AI SDK.
• Managed StackOne Context: The SDK handles StackOne-specific context like Account IDs, API keys, and other required headers behind the scenes. Your agent doesn’t need to worry about managing auth context; it’s handled for you, just as our regular SDK would.
• Lightweight and Dependency-Free: The ToolSet is designed to add as little overhead as possible. The TypeScript package has zero runtime dependencies , so it won’t bloat your agent’s environment.

Try it today: The StackOne ToolSet is available for every StackOne user to experiment with. All you need is a project API Key to start giving your agents tools. Install today via your favourite package manager:

# Install the StackOne ToolSet for Node.js (TypeScript/JavaScript)
npm install @stackone/ai

# Install the StackOne ToolSet for Python
pip install stackone-ai


The code is available on GitHub if you’d like to explore or contribute:

• StackOne ToolSet – TypeScript SDK: stackone-ai-node repository

• StackOne ToolSet – Python SDK: stackone-ai-python repository

Check out the repo, get an API key from your StackOne dashboard and share your feedback. This is an early, experimental release, and your input will help shape the future of Agentic SaaS tools.

StackOne’s mission is to be the universal integration layer for modern software – with Unified APIs for traditional SaaS integration and now tool-calling interfaces for AI agents. With StackOne ToolSet, we’re taking a big step toward that vision. We can’t wait to see the innovative agent solutions you build on top of it. Happy AI hacking!

Want to use our tools in an agent but you’re not a StackOne customer yet? Get a demo or ping us an email at ai@stackone.com

‍