Context & tools
The material the work starts from: the sources, documents and tools a team already relies on.
Updated Oct 7, 2026 · 7 min read
MCP Gateway (October 2026)
TL;DR: An MCP gateway is a control point for Model Context Protocol tool calls: auth, audit, rate limits, observability and policy across your AI fleet. Without one, every agent has its own credentials and failure modes. Swfte’s MCP gateway today checks a per-server API key scope and logs each call. Per-tool policy, SIEM export and call-level rate limits are not built yet.
Six primitives of an MCP gateway, and where Swfte stands
The same primitives that defined API gateways a decade ago apply to MCP gateways. Each card describes the primitive in general, then states what Swfte has built. Where something is not built, we say so.
Tool registry
A centrally managed catalogue of MCP servers that agents and chat clients in the organisation can connect to, with typed, versioned tool definitions.
Swfte today: Partly built. You can deploy MCP servers from templates into a workspace, and each deployment answers tools/list. Versioned tool pinning and a review workflow for new servers are not built.
Auth + secrets
Per-user and per-tool credentials held in a secrets vault, so an agent calling a tool inherits the right identity automatically.
Swfte today: Built at the server level. The gateway requires an API key scoped to that MCP server and checks that the key belongs to the workspace in the URL. Per-user OAuth identity inheritance is not built.
Permission policy
Allow and deny rules per tool and per principal, for example that only Sales agents may call the Salesforce tool.
Swfte today: Not built. Access is decided per MCP server through the key scope, not per tool. Per-tool allow and deny rules are designed for, not shipped.
Audit log
Every tool call logged, ideally with caller identity, arguments, result, latency and cost, retained on policy and exportable.
Swfte today: Partly built. Each gateway call is logged with the workspace, the MCP server, the JSON-RPC method, the status code and the latency, and is mirrored into workspace analytics. It does not record the calling person, the tool arguments or the result, and there is no SIEM export.
Rate limit + budget
Per-tool, per-team and per-user rate limits and call budgets, so a runaway agent hits a cap rather than your invoice.
Swfte today: Not built on the gateway path. Plan limits apply to how many MCP servers a workspace can deploy; per-tool call limits and budgets are designed for.
Observability
Traces, latency percentiles and error rates for every MCP call.
Swfte today: Partly built. Status and latency for each call appear in workspace analytics next to model traffic. Per-call tracing and latency-percentile dashboards for MCP are not built.
How MCP gateway and AI gateway differ
An AI gateway sits in front of LLM providers. The agent or app sends a chat completion request to the gateway; the gateway applies routing policy, rate limits, and audit; the upstream call goes to the model provider.
An MCP gateway sits in front of MCP tool servers. The agent decides to call a tool; the call hits the gateway; the gateway checks the caller’s credential, logs the call and forwards it to the tool server. A fuller gateway also checks per-tool permission and applies rate limits. The tool result returns through the same path.
They are complementary, not competing. Swfte has both: Connect for model calls and a separate MCP gateway endpoint for tool calls. Agents built in Studio can also act as MCP clients and call tools on external MCP servers. Swfte does not currently expose your own agents as an MCP server.
FAQ
What is an MCP gateway?
An MCP gateway is a centralised proxy layer for Model Context Protocol tool calls. Instead of every AI agent or chat client connecting to MCP tool servers directly, they connect through the gateway, which can add authentication, authorisation, audit, rate limiting, observability, and a shared tool registry. A full gateway becomes the single control plane for what tools the organisation's AI fleet can call. How much of that a given product delivers varies, and Swfte's current coverage is listed above.
Why do I need an MCP gateway?
Without one, every agent has its own credentials, its own connection list, and its own failure modes. A gateway gives one set of credentials per tool, one access policy per tool, one audit trail across the fleet, and one place to enforce budgets. The same reasons traditional API gateways exist apply to MCP.
Which MCP gateway should I use?
It depends on what you need to enforce. Open-source MCP proxies are lightweight and suit single-team deployments. Edge and API-gateway vendors have been adding MCP features, so check each vendor's current documentation for what it enforces. Swfte has an MCP gateway in its runtime that checks a per-server API key scope and logs each call, but it has no per-tool policy, no per-call rate limits and no SIEM export today. If you need those controls now, confirm them in a trial before you commit.
Does an MCP gateway work with Claude, ChatGPT, and other models?
MCP is provider-agnostic, so any model or client that supports MCP tool calling can in principle use a gateway. The agent or chat client points at the gateway URL; the gateway checks the credential and forwards the JSON-RPC call to the MCP server.
Is MCP the same as a function calling API?
Related but not the same. Function calling is the per-call mechanism. the model declares what it wants to call, the runtime executes it, the result returns. MCP is the protocol on top, a standard way to describe tools, server capabilities, and resources so any model can discover and use them without bespoke integration code.
Can I write a custom MCP tool server?
Yes: that's the primary use case beyond the public catalogue. The MCP spec is open and SDKs exist for several languages. A custom MCP server typically exposes a handful of tools backed by your internal systems (CRM, ticketing, knowledge base, internal APIs). With Swfte's gateway, access is decided per server by API-key scope, not per tool.
Does an MCP gateway add latency?
Yes. Any gateway adds a network hop plus credential and policy checks. Swfte has not published a measured latency figure for its MCP gateway: <added latency, p95 - founder to fill>.
What is the relationship between an MCP gateway and an AI gateway?
An AI gateway sits in front of LLM providers (Claude, GPT, Gemini). An MCP gateway sits in front of MCP tool servers. They are complementary control planes. Swfte has both: Connect is the model gateway and the MCP gateway is a separate endpoint. They are not yet one shared policy layer; each logs its own calls, and both appear in workspace analytics.
Is MCP secure?
The protocol has an authorization specification, but the security posture depends on the deployment. Without a gateway, every agent holds its own credentials and connects directly to tool servers. A gateway can centralise credential storage, enforce access policy, audit calls and rate-limit abuse; how much a given gateway does is a product question, so ask for specifics. See the MCP security guide for a checklist.
Put your MCP servers behind a scoped gateway
Key-scoped access and call logging for MCP servers deployed in your Swfte workspace. Per-tool policy is designed for, not shipped.
Free tier · Self-hosted deployment on request
Connection map / MCP Gateway
A support agent requests two different tools
Illustrative and designed-for: Swfte's MCP gateway checks a per-server key scope and logs calls today, and per-tool decisions like these are not shipped. A fictional support-team principal. Illustrative authored example · snapshot · 7 September 2026. Supplied catalogue copy, not verified performance evidence; this preview does not execute a workflow.
Request / boundary
support-agent → tickets.read
Routing consideration
The principal is allowed to retrieve assigned support tickets.
Gateway decision: allow the read and record caller identity.
Request / boundary
support-agent → tickets.delete
Routing consideration
The delete tool is outside this team’s approved tool policy.
Gateway decision: deny with the missing-permission reason.
Request / boundary
Reviewer → tool-call record
Routing consideration
Inspect the principal, requested tool and resulting policy decision.
Export record: identity, tool and decision connected in one trace.
How the pieces connect
One useful outcome.
Six deliberate layers.
A spatial explanation of the workflow.
Select a stage or scroll to explore.
EXPLORE THE ASSEMBLY
Intelligence
The reasoning step. A model reads the context that was gathered and proposes something a person can act on.
Review boundaries
The limits set before the work runs — what is in scope, what is refused, and who is asked when it is unclear.
Workflow
The sequence itself: the order of steps, the handoffs between them, and where a person is required.
Deployment
How the result reaches the people who use it, and the conditions under which it is allowed to run.
Evidence
What is kept afterwards so a reviewer can retrace the decision: inputs, choices, and the person accountable.
CONCEPTUAL WORKFLOW MODEL — NOT A DEPICTION OF SYSTEM ARCHITECTURE
Illustrative authored example · snapshot · 7 September 2026. Capabilities and figures require confirmation.