Connect
Optimize
Secure
MCP servers for Vertex AI Agent Builder
MCP Servers
for Vertex AI Agents
Let your Vertex AI agents act across 270+ enterprise SaaS — without leaving Agent Builder. No auth hassle. Token-efficient by design. Security and governance built-in.
What is an MCP gateway?
MCP gateways are real-time integration infrastructure connecting AI
agents to the SaaS apps they need to use. They expose one MCP server
instead of many to the AI agent. They are built on the open Model Context Protocol (opens in new tab), and
are also called MCP aggregators.
MCP gateways use whatever communication protocol each MCP server requires
(stdio, SSE, streamable HTTP), maximize the action coverage agents can
call, and handle authentication, agent action scoping, tool-call
optimization, agent security, and tool-call observability.
StackOne's MCP gateway — SOC2, HIPAA, CCPA, and GDPR compliant — brings this infrastructure to 270+ pre-built SaaS integrations exposing 18,000+ actions, reachable from your Vertex AI agent through one URL.
Act across all your apps
from inside Vertex AI
One MCP gateway, every enterprise SaaS your Vertex AI agent
could ever need to act on.
Coverage
270+ connectors.
18,000+ tested actions.
Every action is built and maintained by StackOne, tested against the live API, and updated when vendors change their endpoints.
Browse all connectors
Trust
Auth & governance,
simplified.
StackOne handles OAuth, API keys, refresh, and scopes for every connector, giving enterprise IT full control through auth configs in a multi-tenant setup.
Agent Auth
Performance
Agent context window,
managed.
Tool Discovery and Code Mode keep your Vertex AI agent's context lean — only relevant actions, no raw response noise. Sharper agents, lower token costs.
Tools Discovery
Security
Agent security,
built in.
Defender scans every MCP tool response for prompt injection in real time. Up to 97.44%¹ detection, 0.2% false positives. SOC2, HIPAA, CCPA, GDPR compliant.
Prompt Injection Defense¹ Jayavibhav test, 65,000 samples.
Connect Vertex AI
to more apps in 4 steps
From auth config to first agent tool call in 5 minutes.
1
Wire StackOne as an MCPToolset in your ADK agent.
In Agent Development Kit (ADK) Python, wrap the StackOne gateway URL in MCPToolset and pass it as a tool on your LlmAgent. Deploy via Agent Engine — available to every agent in your GCP project.
Python — ADK Agent
from google.adk.agents import LlmAgent
from google.adk.tools.mcp_tool import (
MCPToolset, StreamableHTTPConnectionParams,
)
stackone = MCPToolset(
connection_params=StreamableHTTPConnectionParams(
url="https://api.stackone.com/mcp",
headers={"Authorization": f"Bearer {STACKONE_TOKEN}"},
)
)
agent = LlmAgent(
model="gemini-2.5-flash",
name="enterprise_agent",
tools=[stackone],
)
2
Connect each SaaS integration in StackOne.
OAuth or API key per app, done once in your dashboard. Tokens stay server-side.
3
Scope agent actions.
Read-only, specific objects, or per-role exposure. Toggle from the StackOne dashboard.
List records (enabled)
Get record (enabled)
Read related (enabled)
Update record (disabled)
Delete record (disabled)
4
Let your Vertex AI agent call StackOne in plain English.
StackOne maps the agent's tool call to the right action in the right system. Done.
"Sync this week's Salesforce opportunities into BigQuery and open a Linear ticket for any deal blocking the Q3 forecast."
salesforce_list_opportunities
Completed
linear_create_issue
Completed
team: Revenue Operations
priority: high
result: REV-512 created
Synced 38 Salesforce opportunities to BigQuery. Filed REV-512 for the 4 deals blocking Q3.
Get Vertex AI
to do more across your stack
Cross-app workflows your Vertex AI agent now runs in seconds instead of click-throughs.
Customer 360
> Pull Salesforce account context and join it with BigQuery product usage for the next-best action.
HR & Workforce
> Reconcile Workday hires with their GitHub access in seconds.
Finance & Billing
> Reconcile Stripe revenue with the NetSuite GL for month-end close.
Engineering & DevOps
> Open a Linear ticket from a Sentry alert and notify the on-call channel.
Connect any Enterprise AI Agent
to the StackOne MCP Gateway
Same 270+ MCP servers. Same agent context. Pick yours.
Frequently Asked Questions
Vertex AI agents consume external MCP servers through the Agent Development Kit (ADK). In a few lines of Python, wrap the StackOne gateway URL in an
MCPToolset with StreamableHTTPConnectionParams, pass it as a tool on your LlmAgent, and deploy via Agent Engine. 270+ pre-built SaaS connectors are reachable through that single toolset. Full walkthrough in the ADK MCP tools docs. Now part of the Gemini Enterprise Agent Platform.An MCP gateway is real-time integration infrastructure connecting AI agents to the SaaS apps they need to use. It exposes one MCP server instead of many to the AI agent, and handles authentication, agent action scoping, tool-call optimization, agent security, and tool-call observability. Also called an MCP aggregator.
Vertex AI agents can access enterprise SaaS like Salesforce, HubSpot, Workday, Greenhouse, BambooHR, Stripe, NetSuite, Ramp, GitHub, Linear, Sentry, and Datadog through the StackOne MCP gateway — 270+ pre-built connectors in total. StackOne extends Vertex AI beyond Google's first-party MCP services (BigQuery, Maps, Cloud SQL) into operational SaaS. New connectors ship continuously, and you can build custom ones with AI Integration Builder. See all connectors.
Google's first-party MCP servers cover Google-owned data (BigQuery, Maps, Cloud SQL via MCP Toolbox, plus Apigee adapters). StackOne adds the other 270+ operational SaaS — every app reachable through one
MCPToolset instead of one per app, auth managed once in the dashboard, and tool definitions loaded only when relevant to the prompt.Credentials sit at three layers: your StackOne API token (embedded in the gateway URL passed to
StreamableHTTPConnectionParams headers, stored as a Secret Manager secret in your GCP project), SaaS connector credentials (server-side in your StackOne dashboard), and per-user OAuth or session tokens (server-side, scoped per user). StackOne stores OAuth tokens and API keys, refreshes them automatically, and only exposes the actions you've authorized to the agent.Skip the StackOne MCP gateway when any of these is true:
- Google's first-party MCP servers (BigQuery, Maps, Cloud SQL via MCP Toolbox) cover your apps
- You only use a handful of actions across a few apps
- You don't need multi-tenancy, per-user permissions, or per-action scoping
- Indirect prompt injection isn't a concern
- Your MCP provider already does tool discovery and server-side execution to keep the agent context lean
Connect All Your Apps to Vertex AI.
One MCP gateway, 270+ pre-built SaaS connectors. Set up in 5 minutes.