I already have an agent
Connect an MCP client, read answers and receive follow up tasks.
HomeBuild
Build with tools your agent operates.
Pages, forms, bookings, chat and optional meeting memory, through one tool registry, a REST API and an MCP connection.
Three ways to build on it.
I am building an app
Use API keys with specific scopes on your backend, and keep secrets on the server.
I need an internal agent
A hosted pilot reads authorized results and prepares drafts, with activation and credits.
Every product you add is another API to learn.
A link tool, a form tool and a calendar tool, each with its own auth, its own errors and its own idea of permissions.
Without an agent, every time
Learn a new API per productevery product
Handle each one's authseparately
Map errors your agent can act onby hand
Limit what the agent may dorarely possible
Your agent ends up with too much access or not enough tools.
One registry, three surfaces.
The same tools appear in MCP, the REST API and the in app agent, with strict inputs and errors that say what to do next.
- Discovers only the tools its permissions allow
- Gets a hint with every error, and the smallest scope that unlocks it
- Asks the owner for an app or a permission in one request
- Receives events and follow up jobs with receipts
“Connect to instanceof.ai and list what you can do here.”
Connected over MCP with a scoped key
Found the installed apps
Listed the tools this key may use
Ready to workAsk the owner for more access when a task needs it.
Your agent, every product.
- Your agent
- Link page
- Form
- Booking
- Site chat
- Meeting
One connection reaches every installed product, within the scopes you grant.
Built for agents, governed by people.
Strict inputs
An unknown field is an error, never silently dropped.
Only a person installs
Agents request apps and permissions; owners approve them.
Compatibility is honest
It depends on the client's version and configuration. Read the guide before connecting.
What people ask before starting.
Where is the documentation?
In the developer guide, llms.txt for agents, and an OpenAPI description of every endpoint.
Does MCP start my agent?
No. MCP connects tools to an agent that is running. It does not start one.
Can I use it from a backend?
Yes, with scoped API keys kept on the server.