# Connect your own AI agents to Hydden over MCP

Source: https://www.hydden.ai/product-updates/connect-ai-agents-mcp
Last updated: 2026-10-19

> Hydden now runs an MCP server, so your own AI agents can read the identity record and run connector actions within the roles you grant them.

[All product updates](https://www.hydden.ai/product-updates)

AI agents · Integrations · SecurityOct 3, 2026

Connect Claude Desktop, or any agent that speaks the Model Context Protocol (MCP), to Hydden with its own credential, and it sees only the tools that credential is allowed to use.

[Request a trial](https://www.hydden.ai/try-free) [Book a demo](https://www.hydden.ai/book-a-demo)

## On this page

- Overview
- What's new
- How it works
- Getting started

Released

Oct 3, 2026

Availability

All customers

Category

AI agents, Integrations, Security

## Overview

Security teams are building their own AI agents for triage and response, and those agents need identity context: who owns an account, what it can reach and what changed. Handing an agent a broad admin token to get those answers creates the kind of unreviewed access Hydden is there to find.

Hydden now runs an MCP server that any agent can connect to with its own API credential. The agent sees only the tools its roles grant, every action it takes is recorded against that credential, and a suspended credential is refused on its next call.

## What's new

- **Read the identity record:** Accounts, groups, roles, owners, resources and events from every connected system.
- **Trace who can reach what:** Access paths for any account, group, role or resource, including through vaults.
- **Act through your connectors:** Disable an account or run another connector action you have allowed.
- **Bound by its roles:** Each agent sees only the tools its credential may call.

## How it works

An administrator creates an API credential for the agent and gives it roles to read the identity record, view access reviews or run connector actions. Grants can be as narrow as one action on one connector, so a credential allowed to disable accounts can never enable one. Role changes reach a running agent within seconds, each credential has its own rate limit, and every action lands in that credential's audit trail.

The grant path an agent can ask for, from an account through a nested group to a resource.

Cutting an account's access, one of the connector actions a credential can be allowed to run.

## Getting started

The MCP endpoint is part of Hydden Control for all customers. Create an API credential for your agent, assign its roles and add Hydden to the agent's MCP servers.

**Need configuration detail?** Technical documentation is available to customers after sign-in.

## Related updates

[All updates](https://www.hydden.ai/product-updates)

- Search and reporting · SecuritySep 1, 2026 **[Ask the AI assistant who can reach what](https://www.hydden.ai/product-updates/ai-assistant-access-questions)** Ask what a group grants, who can reach a vault and which accounts look over-privileged.
- IntegrationsOct 3, 2026 **[Guided connector setup and a public API](https://www.hydden.ai/ecosystem)** New connectors open into a four-step checklist, and integrations can use the documented API.
- NewAI agents · SecurityOct 3, 2026 **[AI agent profiles and blast radius](https://www.hydden.ai/product-updates/ai-agent-profiles-blast-radius)** See every AI agent, what it has done, and how much any identity can reach.

## Give your agents the identity record

Request a trial and connect your first agent to Hydden.