Is an MCP gateway the same as an AI gateway?

They overlap. An AI gateway usually sits in front of model providers; an MCP gateway sits in front of MCP servers and understands tool calls. Some products do both.

Does an MCP gateway see local MCP servers?

No. A local MCP server started by the agent on the same laptop talks to it directly, usually over stdio, so the traffic never passes a network gateway. On-device control covers that case.

Which products here describe an MCP gateway?

On the pages we read: Harmonic Security, Lasso Security (open source) and Zenity, with Onyx Security listing gateways and APIs more generally. Cloudflare AI Gateway sits in front of model providers rather than MCP servers.

CONTROL TYPE · GWY

AI and MCP gateways: one checkpoint for model and tool calls

Short answer

An AI or MCP gateway routes calls to models and MCP tools through one checkpoint that can filter, block, rewrite and log them. It works without touching the device, which makes it useful for cloud workspaces and devices you do not manage. It sees only the calls routed through it: a local MCP server talking to an agent on the same laptop never reaches it.

By The Device Desk · Endpoint AI Finder · Published 2026-10-11

Desk research from public vendor pages read on 27 September and 11 October 2026. Profiles follow our published method; we did not run the products.

Vendor pages read September and October 2026 · updated 2026-10-13

What does AI or MCP gateway do?

A gateway sits between AI clients and what they call: a model provider or a remote MCP server. Every call that passes through can be checked against policy, changed, refused or logged.

MCP gateways are the newer kind. They know the MCP protocol, so they can see which tool is being called with which arguments. The 2026-07-28 MCP specification added Mcp-Method and Mcp-Name headers to Streamable HTTP requests, which the release post says lets gateways and similar intermediaries route and meter traffic without parsing the request body.

Five products describe a gateway on the pages we read. Lasso Security offers an open-source MCP gateway that can block, alert or sanitize. Zenity's MCP gateway can block or modify a tool call before it executes. Harmonic Security lists MCP gateways among its surfaces, and Onyx Security lists gateways and APIs. Cloudflare AI Gateway is a different kind: it sits in front of the model providers your own AI applications call, with analytics, logging, caching, rate limiting, retry and model fallback, and it does not see employees' tools. Palo Alto Networks completed its acquisition of the AI gateway company Portkey on 29 May 2026 to serve as the gateway for Prisma AIRS, a separate product from Koi.

When do you need it?

Work happens in cloud workspaces, or on devices you cannot install anything on.
Teams use remote MCP servers and you want one place to approve and log them.
You build your own AI apps or agents and want calls to models routed through policy.

What can it not see?

A gateway sees only what is routed through it. Local MCP servers that talk to the agent over stdio, file reads, shell commands and anything a user sends through a tool that bypasses the gateway are invisible to it. Routing has to be enforced, usually by configuration on managed devices or by network rules, or people can go around it.

Which finder answers point to GWY?

+2if the answer to "Which AI tools do employees use at work?" is Coding agents and MCP servers
+1if the answer to "Do you manage the devices people use for AI?" is A mix of managed and unmanaged
+2if the answer to "Do you manage the devices people use for AI?" is Mostly unmanaged devices
+1if the answer to "Where does the work happen?" is Partly in cloud workspaces
+3if the answer to "Where does the work happen?" is Mostly in cloud workspaces or virtual desktops
+1if the answer to "How sensitive is the data AI tools can reach?" is Regulated data, secrets or production access

Try the finder

Which products describe AI or MCP gateway?

5 of 19 products describe AI or MCP gateway on the pages we read, 1 partly, and 13 are not in our sources.

Alphabetical. From each product's own pages.
Product	Status	What their pages say	Source
AI Access Security (Palo Alto Networks)	Not in our sources	Not described on the pages we read.	None
Akamai Workforce Protector	Not in our sources	Not described on the pages we read.	None
Bay	Not in our sources	Not described on the pages we read.	None
Bloom Security	Not in our sources	Not described on the pages we read.	None
Cloudflare AI Gateway	Described	An AI gateway with analytics, logging, caching, rate limiting, retry and model fallback for AI applications.	Source: Cloudflare AI Gateway documentation · Read 11 Oct 2026
CrowdStrike Falcon Guardian	Partly described	The AI gateway is listed as coming soon; SACR (secondary) also lists an MCP proxy.	Source: CrowdStrike: Falcon Guardian (formerly AIDR) · Read 13 Oct 2026
Ent	Not in our sources	Not described on the pages we read.	None
Glow Security	Not in our sources	Not described on the pages we read.	None
Harmonic Security	Described	MCP gateways are a listed surface, and Harmonic Command governs at the MCP layer and at the tool surface.	Source: Harmonic Security: Command · Read 27 Sep 2026
Koi (Palo Alto Networks)	Not in our sources	Not described on the pages we read.	None
Lasso Security	Described	An open-source MCP gateway that discovers MCP servers across Claude, Cursor and Windsurf and can block, alert or sanitize.	Source: Lasso Security: MCP security · Read 27 Sep 2026
Neo	Not in our sources	Not described on the pages we read.	None
Noma Security	Not in our sources	Not described on the pages we read.	None
Onyx Security	Described	Gateways and APIs inspect traffic inline.	Source: Onyx Security: AI security platform · Read 27 Sep 2026
Pluto Security	Not in our sources	Not described on the pages we read.	None
Prompt Security (SentinelOne)	Not in our sources	Not described on the pages we read.	None
Sun Security	Not in our sources	Not described on the pages we read.	None
Zenity	Described	The Zenity MCP gateway can block or modify a tool call before it executes.	Source: Zenity: coding and personal agents · Read 27 Sep 2026
Zscaler Generative AI Security	Not in our sources	Not described on the pages we read.	None

Not in our sources is not a no. Ask the vendor, and send us a public source if one exists.

What should you ask?

Which MCP servers and model providers can you sit in front of, and how is traffic forced through you?
Can you block or modify a single tool call, not only a whole server?
Do you support the MCP 2026-07-28 revision, and what happens with servers on older versions?
What do you see of local MCP servers on the laptop?
Can you tie each call to the person and the agent that made it?

Sources

Source: MCP blog: the 2026-07-28 specification · Read 27 Sep 2026

Source: Lasso Security: MCP security · Read 27 Sep 2026

Source: Zenity: coding and personal agents · Read 27 Sep 2026

Source: Harmonic Security: Command · Read 27 Sep 2026

Source: Onyx Security: AI security platform · Read 27 Sep 2026

Source: Cloudflare AI Gateway documentation · Read 11 Oct 2026

Source: Palo Alto Networks press release: acquisition of Portkey completed · Read 27 Sep 2026

Learn more

AI on devices you do not manage: what you can still control
None of the products we cover describes unmanaged-device coverage. That is not a gap in one vendor; it is how device controls work.
AI controls for cloud workspaces and virtual desktops
When the work runs on a server, the endpoint that matters is the server, not the laptop.