Skip to content
Halo by OmniTrust

Secure · MCP and Tool Gateway

Available now

Put policy in front of agent tool calls

Halo's MCP and tool gateway sits between agents and the tools they call. It applies identity, policy, approvals, budgets, and evidence to tool traffic, and it brokers credentials so agents never see raw keys.

Book a walkthrough Start with Halo
separate gateways, model, tool, and dispatch
3
apps in the connection catalog
1,000+

Why it matters

Tool access needs a checkpoint

The Model Context Protocol gives agents a standard way to reach tools and data, and each connected tool widens what an agent can do. Without a control point, the agent's own configuration decides which calls happen and which keys it holds.

  1. 01

    Tool access granted at setup time cannot account for what an agent attempts at run time.

  2. 02

    API keys placed in agent configuration can leak through logs, prompts, or model output.

  3. 03

    Reviewers cannot see which tool calls were allowed, denied, or held unless something records them.

How it works

What happens when an agent calls a tool

The path is short, and each stage leaves a record.

  1. 1

    The agent calls a tool

    The agent runtime routes the tool call to the Halo gateway along with the agent's identity and the call details.

  2. 2

    The gateway evaluates

    The gateway checks identity, evaluates the policy bundle against the tool, arguments, and destination, and applies budget limits.

  3. 3

    Policy decides

    Policy returns allow, deny, or hold for approval. Held calls wait for a reviewer, and denied calls return an error to the agent.

  4. 4

    Broker and record

    For allowed calls, the gateway attaches the stored credential, forwards the call, and records the decision as evidence.

Capabilities

Controls where agents meet tools

The gateway evaluates agent traffic before it reaches a tool, then records what it decided.

  • Agent identity checks

    The gateway resolves which agent is calling and applies access gates that deny by default, so a call without a recognized identity does not pass.

  • Tool-level policy

    Policy bundles evaluate the tool, its arguments, and its destination, then return allow, deny, or hold for approval.

  • Credential brokering

    Provider and app credentials stay bound at the gateway. Agents receive scoped capabilities and never see raw keys.

  • Human approvals

    Higher-impact calls can pause for a person to approve. The run step waits, and the reviewer's decision is recorded.

  • Budget checks

    Per-agent budgets are checked at the gateway, so spend limits apply to the traffic that creates the cost.

  • Evidence for decisions

    Gateway decisions record the policy, the outcome, and the reason, and they export as CSV and JSON for review.

Credential brokering

Credentials stay at the gateway

Agents that hold raw API keys can leak them through logs, prompts, or output. With credential brokering, Halo stores the credential, binds it at the gateway, and gives the agent a scoped capability for the approved work.

  • Credential values are write-only in Halo
  • Agents receive scoped capabilities
  • Gateways attach credentials at call time
  • Credential use is recorded as evidence

FAQ

Questions, answered

An MCP gateway sits between AI agents and the tools they reach through the Model Context Protocol, an open protocol introduced by Anthropic in November 2024. Halo's MCP and tool gateway applies identity, policy, approvals, budgets, and evidence to that traffic.

Halo uses separate gateways. The model gateway handles calls to language models, and the MCP and tool gateway handles calls to tools and apps. Both apply identity, policy, approvals, budgets, and evidence to their traffic. A separate API and dispatch gateway fronts how work reaches agents.

No. Halo stores provider and app credentials and binds them at the gateway. Agents receive scoped capabilities and never see raw keys. Credential values are write-only, so they cannot be read back from the Halo interface after you save them.

Tools you have connected and approved. The Halo connection catalog lists 1,000+ apps, and available actions vary by app. Policy decides which tools and arguments an agent may use, and approvals can hold higher-impact actions for a person.

The gateway returns an error to the agent, and the denial is recorded as evidence with the policy and the reason. In audit mode, the match is recorded and the call proceeds, which lets you test a rule before you enforce it.

Related capabilities

Model GatewayManaged and BYOK model accessLearn more about Model GatewayApp Connections1,000+ apps in the connection catalogLearn more about App ConnectionsHuman ApprovalsPut people on high-impact actionsLearn more about Human Approvals

Put a halo around the AI your business runs.

See discovery, runtime policy, approvals, and evidence working together on your own use cases.

Book a walkthrough Start with Halo