Advanced

MCP and custom tools for AI agents

An AI agent calls your specialized tools while business rules, permissions and execution remain in a controlled server-side flow.

Browser call

A workshop enquiry ends with a reserved place

The AI agent uses a custom connection to your workshop system to read availability and register a participant. This flow requires connected schedule checks and a booking action.

Circle Studio Browser call

Call transcript

  1. Customer

    I would like to book the beginner pottery workshop.

  2. Telp

    There are two places on Saturday at 11 am. It lasts two hours and costs RSD 3,500, payable on arrival. Please give your name and email to book.

  3. Customer

    Book one place for Saturday at 11 am. Ana Example, [email protected].

  4. Telp

    Ana, your place is reserved for Saturday at 11 am, reference T-807. The workshop lasts two hours and you pay RSD 3,500 on arrival.

Telegram

A new website enquiry becomes a project task

The AI agent clarifies scope, deadline and contact. A custom connection creates and assigns the request in your project system; the team prepares the estimate and quote.

Forma Studio Telegram
  1. Customer

    We need a company website in Serbian and English. Could you send a quote?

  2. Telp

    How many pages and what deadline do you have in mind? Please include the company name, your name and email for the quote.

  3. Customer

    Six pages, by 1 December. Example Ltd, Ana Example, [email protected]. Please send the request to your team.

  4. Telp

    Ana, task P-808 is created and assigned to the project team: six pages in Serbian and English for Example Ltd, requested by 1 December. The quote will go to [email protected].

The customer context

An AI agent calls your specialized tools while business rules, permissions and execution remain in a controlled server-side flow.

Expose specialized operations when the business logic and authorization should remain owned by your server.

What this workflow can cover

  • Register narrowly described server-owned operations
  • Keep lifecycle and authorization decisions outside the model
  • Reuse tool contracts across compatible agent experiences
  • Audit execution outcomes through platform surfaces

Advanced

How it works

  1. 1

    Model each operation as a typed, permission-aware tool.

  2. 2

    Attach it only to the agents and workflows that need it.

  3. 3

    Validate behavior across chat and voice before production rollout.

What your team gets

Stronger governance

Reusable business contracts

Less hidden behavior in prompts

Telp

How your team keeps control

Credentials remain server-side

Secrets are not placed in prompts or customer-visible conversation text.

Every tool has a contract

Names, descriptions, fields and permission policy define what the agent may request.

Writes can require confirmation

The platform can prepare a structured draft and execute only after explicit approval.

FAQ

Frequently asked questions

When does an AI agent need custom tools?

When your process requires a specific check or action not covered by an existing connection, such as checking capacity by location or creating an internal project task.

Does connecting MCP tools give the agent every capability of our system?

No. You choose which tools the agent may use and which data it may send. Business rules and access checks remain in the system executing the action.

Can the same business tool be used in chat and voice?

Yes, when the action is supported in both experiences. Confirmation needs separate checking: a tool requiring written confirmation does not execute based only on verbal approval.

Who determines whether a custom action completed successfully?

The connected system returns the execution result, and Telp records the outcome. The agent must not confirm a reservation, entry or delivery if the action failed.

Want to validate this workflow with your own examples?

Tell us the channel, current process and first result you want to improve. We will suggest a focused rollout path and the cases your team should test.

View pricing