Operations team connecting an AI agent to an internal REST API
Point the agent at any API endpoint and it calls it as part of its workflow. No custom integration code.
Let Your Employee Call Any API
Give an employee a config for any REST or GraphQL endpoint, the method, headers, auth, and the shape of the request and response, and it will become a tool the employee can call during a task. Unlike a one-way webhook trigger, this reads data back: the employee gets a structured response it can reason about and act on, not just a fired event. No code required on your side. This is on the roadmap and is not available yet.
Not every tool has an OAuth integration or MCP server. Custom API endpoints cover the rest. Your employee calls any REST or GraphQL endpoint, sends payloads, reads responses, and chains multiple API calls together into workflows.
API connections support custom headers, authentication tokens, and request templates. Define the endpoint once, and your employee can call it whenever the task requires it. No code, no deployment, just configuration.
This is the universal fallback. If a system has an API, your employee can use it. Internal tools, legacy systems, custom microservices, third-party platforms without OAuth support. Configure the endpoint, and your employee treats it like any other tool.
API endpoint tools are on the roadmap and are not configurable yet. What is coming is a form rather than a codebase: a name, a plain description of what the endpoint does, the address, the method, the headers, and the shape of the request. The employee reads that description the way it reads every other tool description and works out on its own when calling it is the right move.
The part that matters is the return trip. A webhook fires an event at another system and moves on, which is exactly right when the other side only needs to be told something happened. An API endpoint tool waits for the answer and hands the employee a structured response it can reason about, so it can ask your inventory system how many units are left and then use that number in the sentence it is writing to a customer.
Two paths already cover a lot of this ground while the endpoint tool is on its way. Anything with an official integration connects through the app catalog in one click, and anything your own team runs can be published as an MCP server, which employees discover and call today with no code needed on our side. The endpoint tool is for what is left over: a system that has an API, has no integration, and has nobody free to wrap it.
Turn Any API Into a Tool Your Agent Can Use
Sistava lets you configure any REST or GraphQL API as a tool your AI employee can call during a task. Define the endpoint URL, HTTP method, headers, authentication, and request template, and the agent gains the ability to call that API whenever it needs to.
This means any internal system, any third-party service, or any custom backend your company has built can become part of your AI employee's toolkit. No webhook, no OAuth flow required. If it has an API, your agent can use it.
Full Control Over Authentication and Request Shape
API endpoint integrations support custom headers, bearer tokens, API keys, and request body templates. You define exactly what the agent sends and how the API expects to receive it. The agent fills in the dynamic parts based on context at call time.
This level of control means you can connect to APIs with strict authentication requirements, custom content types, or non-standard request formats without any workarounds. Your agent calls the API exactly as it expects to be called.
Useful for Internal Systems and Custom Backends
OAuth covers public apps, but most companies also have internal APIs: ERP systems, custom data services, internal reporting endpoints, legacy backends. API endpoint integrations cover everything OAuth does not.
You can also use API endpoints to compose multi-step workflows where the agent calls a series of internal services to complete a task. Each API becomes a discrete tool, and the agent decides the order and logic of execution based on the goal you give it.
How It Works
Your AI agents call any REST or GraphQL API on demand by using endpoint definitions you configure once and never touch again.
Define an API endpoint by providing the URL, HTTP method, headers, authentication, and a description of what it does. Your AI employee reads the description and decides when and how to call it based on what a task requires. You do not map triggers or write wrapper code. The agent reasons about the endpoint the same way it reasons about any other tool in its kit.
Custom headers, bearer tokens, API keys, and basic auth are all supported. The agent handles parameter substitution, request construction, and response parsing automatically. This makes any internal microservice, legacy backend, or third-party API without an OAuth connector available to your entire AI workforce. Define it once and every agent that needs it can use it.
Credentials live on the endpoint configuration rather than in the employee's hands, the same way connected apps work today, so a raw key is never part of what the model sees. The endpoint sits in the Tools tab beside every other capability and carries the same approval setting, which means anything that writes to a live system can be made to stop and ask you first. Every call it makes shows up in the activity log with what was sent and what came back.
Use Cases
Finance team giving an AI employee access to live pricing data
The agent fetches current rates from an external API and uses them in every calculation.
Support team connecting an AI agent to a ticketing system's API
The AI employee reads, updates, and creates tickets through the API. Every action stays in sync with the real system.
Developer testing an AI agent against a staging API environment
Configure the endpoint, set the auth, and the agent is making calls immediately. No deployment required.
Comparison
| Before | After |
|---|---|
| Connecting an agent to a custom API requires engineering time and maintenance. | Configure any API endpoint directly in the platform and the agent calls it. |
| The agent works with stale data because it can't reach live systems. | Real-time API calls give the AI employee access to current data. |
| Every new data source means a new integration project. | Add an endpoint, the agent uses it immediately. |
| Custom integrations are brittle and hard to update. | API endpoint configuration is centralized and version-independent. |
FAQ
What API formats does Sistava support for custom integrations?
Sistava supports both REST and GraphQL APIs. You can configure any HTTP-based endpoint, including GET, POST, PUT, and DELETE methods, with custom headers and authentication.
Can I use API keys or bearer tokens to authenticate the calls?
Yes. You can define authentication headers directly in the integration configuration. The credentials are stored securely and injected into each API call automatically. The agent never has direct access to the raw credentials.
How does the agent know when to call a custom API endpoint?
Each API endpoint is given a name and description when you configure it. The agent uses these to understand what the tool does and decides when to call it based on the task at hand, the same way it decides when to use any other tool.
Can I connect to an internal API behind a firewall?
For internal APIs accessible over HTTPS, yes. If your internal API requires network-level access, you would need to expose it via a secure tunnel or VPN endpoint. The platform makes the outbound call, so the API endpoint needs to be reachable from Sistava's infrastructure.
Can my AI employee call our internal APIs or custom backend services?
Yes. You can register any REST or GraphQL endpoint and the AI agent will call it as part of completing tasks. Authentication, headers, and parameters are all configurable.
Is the custom API endpoint tool available yet?
Not yet. API endpoint tools are on the Sistava roadmap and cannot be configured today. If you need an external system reachable right now, connect it through the app catalog if it has an integration, or publish it as an MCP server, which employees can already discover and call.
Can I connect Sistava to a tool that has no official integration?
Yes, in two ways. Anything reachable over HTTPS can be published as an MCP server and your employees will discover its tools and call them today. The API endpoint tool is the second route, arriving so you can point an employee at a plain REST or GraphQL address with no server to build at all.
What is the difference between an API endpoint tool and a webhook?
A webhook is one directional: it fires an event at another system and the employee moves on without an answer. An API endpoint tool waits for the response and hands the employee structured data it can reason about, which is what you need when the employee has to use the result rather than just announce something.
We exposed our inventory system as an API endpoint and the support agent started pulling real stock data into customer replies. No code changes on our end.
Where Let Your Employee Call Any API fits
Let Your Employee Call Any API is part of Things they connect to.
Connect your AI agents to Gmail, Slack, Notion, Jira, HubSpot, and thousands more apps with one-click OAuth. Extend their reach with MCP servers and A2A agents for AI-to-AI interoperability, or call any REST or GraphQL API directly. Each connection is scoped per employee.
- Things they connect to: 1,000+ tools. One click to connect.
Read the guide
More in Integrations
- 1,000+ App Integrations: Connect your AI employee to Gmail, Slack, Notion, HubSpot, and over a thousand other apps through secure OAuth, no API keys required. Authorize an app once for your company, then switch it on for any employee with one click. Your employee takes real action inside these apps, sending the Slack message, updating the HubSpot record, drafting the Gmail reply, instead of just describing what it would do.
- E-commerce Management: Manage a Shopify store with an AI employee that reviews catalog, inventory, orders, and market evidence, then asks you before every live store change. Other e-commerce connections are catalog-listed and stay review-first while their provider actions are verified.
- Give Your Employee Any MCP Server: Paste the URL of any external MCP (Model Context Protocol) server into an employee's Tools tab, and Sistava discovers every tool it offers and hands them to that employee, with no code required. The connection is granted to the employee's whole team, so a database server, an internal API, or a community-built MCP server becomes a tool your team can call by name in every conversation. This is the fastest path to extending an employee past the built-in tool set.
- Give Your Employee an External AI Agent: Point an employee at any external A2A (agent-to-agent) agent's URL and it becomes a tool the employee can call on its own. Sistava fetches the agent's public Agent Card, turns each of its skills into a usable tool automatically, and keeps them refreshed as the remote agent changes. No manual mapping of inputs or outputs required.
- Let Your Employee Trigger Webhooks: Point an employee at any outbound webhook, your own automation, a Zapier or Make scenario, an internal service, anything that accepts an HTTP call, and it becomes a tool the employee can trigger on its own during a task, not something you fire manually from a dashboard.