Dome Systems

Use case

Consolidate tool access

Every agent reaches for tools: CRMs, ticketing, file stores, custom MCP servers, internal APIs. Today those credentials sit in the agent, and trust ends with the client. A single gateway over every call is the right boundary, where access is limited, policy enforced, and every action audited.

Agentic patterns

Tool calls happen on both sides of the wire

The same gateway primitive applies whether you're governing the tool calls your agents make, or exposing your APIs as a tool surface for customer agents to call. The control shifts; the boundary stays the same.

AI In

Your internal estate

The tool calls your agents make against CRMs, ticketing, file stores, custom MCP servers, and internal APIs. One gateway replaces credentials in agent code and per-tool re-implementations.

Scenario

A support-triage agent reaches for Salesforce, Jira, Snowflake, and a custom MCP server. Without a gateway, credentials live in each agent and audit lives in four different formats.

Outcomes that matter

  • One authentication model across every tool
  • Policy at the call boundary, with full context
  • One audit trail across every action

AI Out

Your customer and partner estate

Customers and integration partners reach for your APIs through agents. An MCP tool surface in front of those APIs gives you the operational discipline agents need.

Scenario

A customer's agent drives your billing API with a sequence of calls you didn't anticipate. You need intent-aware policy and per-customer attribution, not just per-key throttles.

Outcomes that matter

  • Stand up an MCP surface without rewriting your APIs
  • Per-customer attribution, rate limits, and quotas
  • Intent-aware policy alongside the API gateway

The tool call boundary

Tool calls are where governance actually happens

Authorizing the agent is necessary but not sufficient. The call itself carries arguments your policy was never designed for. The gateway is where you control what the caller is trying to do, not just whether they're allowed to call.

Agent

Tool call request with intent

Identity

Agent + user identity verified

Policy

Evaluate args, allow/deny/judge

Tool

Credential injected at boundary

Response

Filtered, audited, returned

One gateway in front of every tool means one place to verify the caller, one place to evaluate policy, and one place to audit. No credentials in agent code. No tool-by-tool re-implementations of the same controls.

What you get

One boundary, three operations

Limit access

Constrain the systems and data each agent can reach. Tools attach to agents explicitly; the gateway refuses anything unattached. No standing credentials in agent code.

Enforce policy

Every call evaluated against policy with full context: agent identity, user identity, tool, arguments. Allow, deny, or escalate. Outputs filtered before they return.

Audit every call

Each tool call recorded with who, what, under which rule, and with what result. Queryable. Streamable to the SIEM and SOAR systems your team already operates.

Before and after

Where the operational work ends up

The work doesn't disappear; it relocates. Per-tool, per-agent, per-team controls collapse into one place that the platform team operates and security can reason about.

Before

After

Distributed control

Credentials live in each agent. Each tool has its own auth model. Audit is per-tool and per-format. Policy decisions are duplicated across agents that all call the same backend.

One gateway

Credentials live in the gateway. Policy is attached to the tool, not the agent. Audit is one event vocabulary across every action. Adding a tool is a configuration, not a refactor.

Output trust

Whatever the tool returns flows back to the agent. PII, restricted records, credentials embedded in payloads, all of it landing in the agent's context window.

Output governance

Responses pass back through policy. Fields can be redacted, records refused, sensitive substrings filtered. Outputs governed as well as inputs.