Every major integration platform now says it supports the Model Context Protocol. That is the easy part: wrapping an API as an MCP tool takes an afternoon. The hard part is what happens on the hundredth agent and the ten-thousandth call. Whose credentials did the agent use? Was that action within its job? Can security see it? Unmanaged MCP is risky: industry research cited in our own field material puts the exploit success rate on unmanaged MCP servers at 92%.
So “does it support MCP” is the wrong question. The right one is whether the iPaaS gives you a governed MCP layer on top of the systems it already connects. Here is what native MCP support should mean, and how the major platforms approach it.
What native MCP support means
An iPaaS with native MCP support does three things on the same foundation it uses for integrations:
- Publishes tools. Connectors and workflows become MCP tools that agents in Claude, ChatGPT, Copilot Studio and other MCP clients can call, without a separate server per app.
- Governs calls. Each call is authenticated, tied to a user, scoped to what the agent is allowed to do, and logged.
- Reuses what you already run. The same connectors, credentials, error handling and monitoring your integrations use, rather than a second stack just for agents.
If a platform’s MCP layer lives outside its integration engine, with its own auth and its own logs, you get a second system to secure. That is the gap native support closes.
Six checks before you buy
1. Can it turn any connected system into a tool? Prebuilt MCP servers for popular apps are useful, but agents need your long tail too: internal APIs, niche SaaS, databases. Check whether any connector, or any REST API, can become a governed tool.
2. Is every call tied to a named user? An agent acting on a shared service account is an audit gap. Look for per-user identity, ideally running on the user’s own credentials so the agent can never do more than the person could.
3. Can you scope tools to the job? An agent that only needs to read invoices should not be able to delete them. Check that you can expose specific actions, not whole APIs.
4. Is there one audit trail? Security needs one place to see which agent called which tool, for whom, with what result, across every app.
5. Can it package multi-step work into one tool? Raw APIs make agents chain many calls, and every tool definition and result costs tokens. Composite tools do a multi-step job in one call, which is cheaper and more predictable.
6. Can it govern MCP servers you did not build? Teams will bring their own MCP servers. Check whether the platform can register, proxy or govern third-party servers under the same controls.
How the major integration platforms support MCP
We read each vendor’s own pages on 8 November 2026. For the full platform comparison, including pricing and where each one runs, see our enterprise iPaaS comparison.
| Platform | How it supports MCP |
|---|---|
| Tray.ai | Agent Gateway turns Tray connectors into governed MCP tools. Each call is tied to a named user, can run on their own credentials, is scoped to its job and is logged in one audit log. Composite tools do a multi-step job in one call. |
| Workato | Enterprise MCP turns recipe functions and skills into MCP servers, with production-ready servers for apps such as Salesforce, Slack, Jira and Workday, and agent access through a unified gateway. |
| Celigo | Its MCP server exposes approved Celigo tools, custom APIs and app connections to MCP-compatible agents, and can proxy upstream MCP servers under Celigo’s governance controls. |
| Boomi | Boomi Connect offers over 1,000 prebuilt MCP servers with OAuth pass-through that keeps each user’s own permissions, and its AI Gateway covers AI and MCP interactions. |
| MuleSoft | Agent Fabric runs model calls, MCP tool access and agent-to-agent traffic through Omni Gateway under one policy set and audit trail, and existing APIs can be mapped to MCP tools. |
| SnapLogic | Works in both directions: pipelines give agents managed access to applications, and an MCP Client Snap Pack lets pipelines call remote MCP servers. Its AI Gateway adds authentication, rate limits and audit controls. |
| Jitterbit | Creates, exposes and manages MCP servers without custom code, with gateway protections, identity management and message inspection. |
| Informatica | Informatica Headless exposes data management, such as pipeline building, data quality and MDM, through MCP servers, APIs and agent-to-agent services. |
| Microsoft | Copilot Studio agents use tools from MCP servers, and Logic Apps Standard can expose workflows as remote MCP servers, a feature in preview. |
Where Tray.ai fits
Tray.ai connects to 12,000+ apps, databases, APIs, and services, and Agent Gateway for MCP can publish any of them as governed MCP tools. Integrations, workflows and agents run on the same platform with the same connectors, so the controls you apply to integrations apply to agents too.
- Per-user identity: each call is tied to a named user and can run on their own credentials.
- Scoped tools: expose the actions an agent needs, not the whole API.
- One audit log across every tool and every app.
- Composite tools: a multi-step job in one call, for up to 10x lower token costs than raw tool chains.
To compare dedicated MCP gateways as well as integration platforms, see the best MCP gateways, or read how MCP compares with calling APIs directly. If you want to try it on your own systems, start a free 30-day trial.