Kong Konnect MCP Server

The Kong Konnect MCP Server connects AI assistants to Kong Konnect, Kong's cloud platform for API gateways, developer portals and API analytics. If you run your APIs through Kong Gateway, you can use it to query services, routes, consumers and plugins straight from your editor, analyse traffic and narrow down errors without opening the Konnect interface. MCP, the Model Context Protocol, is an open standard. Through it, an assistant such as Claude Code, Cursor or GitHub Copilot calls external tools. Kong runs the server itself: according to the documentation it is hosted by Kong and talks to the Konnect backend services behind the scenes. There is nothing for you to install or upgrade.

How the server is provided

The server lives at a single global address, https://global.mcp.konghq.com/, no matter which Konnect region your resources sit in. Kong also lists regional endpoints for the US, Europe, Australia and India, such as https://eu.mcp.konghq.com/. That helps teams that deliberately keep data paths inside one region. According to Kong, queries respect regional data boundaries and the organisation's role-based access controls. MCP access is switched on by default for every organisation. Administrators can turn it off in AI Settings, after which nobody in the organisation can reach the server.

Kong has offered the server as a public beta since 19 December 2025. According to the release notes, the standalone MCP Code Mode became generally available on 18 May 2026. The older open-source project Kong/mcp-konnect on GitHub is archived and itself points to the hosted server, so the offering on konghq.com is the one that counts today.

Signing in: browser, personal token or service account

There are three ways to authenticate. In clients that support it, including Claude Code and VS Code, you simply sign in to Konnect in the browser and authorise access. Alternatively, you create a Personal Access Token (PAT) in Konnect and pass it as a bearer header. For automated, unattended workflows Kong provides a System Account Access Token, which is tied to a service account with defined permissions. In every case, the assistant only sees what the underlying account is allowed to see.

What can the Kong Konnect MCP Server do?

Instead of dozens of individual tools, the server follows a "discover, then execute" pattern. The search tool finds the right Konnect API operation from a plain description, such as "list all services in a control plane". With get_schema the assistant fetches the parameters it needs, and execute finally calls the operation. According to Kong, this lets the server cover a large part of the Konnect API without flooding the model's context with long tool lists.

Kong describes five functional areas. First, managing gateway entities: control planes, services, routes, consumers, consumer groups, plugins and vaults. Second, debugging: the assistant can start debug sessions with active tracing to follow slow or failing requests. Third, analytics: API requests can be filtered by time range, status code, consumer, service and route. Fourth, searching Kong's documentation through the KnowledgeBase tool, which is only available when the KAi assistant is enabled. Fifth, maintaining the Dev Portal: creating, editing and publishing pages, snippets and Markdown components.

Costs and requirements

The documentation lists no separate price and no plan restriction for the MCP server itself. You need a Konnect account with suitable permissions. On its pricing page, Kong lists a Plus tier with usage-based billing and an Enterprise tier with custom quotes. Plus plans come with a 30-day trial that needs no credit card; the page does not clearly offer a permanently free tier. Current terms are on Kong's pricing page.

Who is the Kong Konnect MCP Server for?

It pays off for platform, API and DevOps teams that run Kong Gateway through Konnect. Typical questions include "Which routes are attached to this service?", "Why has this consumer been getting 429 errors since this morning?" or "Which plugins are active on the production control plane?". The assistant answers questions like these in seconds. If you only run Kong Gateway yourself without Konnect, it offers you nothing, because the server talks exclusively to the Konnect APIs. Since execute can also call write operations, start with a token that has as few rights as possible and have the client confirm each change individually.

A Kong Konnect account with suitable permissions, MCP access not disabled for the organisation, network access to global.mcp.konghq.com or a regional endpoint, and an MCP client with HTTP transport, such as Claude Code, Cursor, VS Code with GitHub Copilot or Windsurf.

With the Kong Konnect MCP Server, an AI assistant queries gateway configuration, analytics and Dev Portal content straight from Kong Konnect. This guide takes you from setup to your first query.

Step 1: Prerequisites

You need a Kong Konnect account with rights on the control planes you want to query, and an MCP client with HTTP transport, such as Claude Code, VS Code, Cursor or Windsurf. According to Kong, MCP access is switched on by default for every organisation. Your firewall must allow connections to global.mcp.konghq.com.

Choose an endpoint

The global address https://global.mcp.konghq.com/ works regardless of where your resources live. If you want to keep data paths inside one region, use a regional endpoint such as https://eu.mcp.konghq.com/ or https://us.mcp.konghq.com/ instead; Kong also offers au and in.

Step 2: Connect it to Claude Code

With browser sign-in, the command is claude mcp add --transport http kong-konnect https://global.mcp.konghq.com. Claude Code then opens a browser window where you sign in to Konnect and authorise access. With a token, append --header "Authorization: Bearer YOUR_KONNECT_PAT". Then run claude mcp list to check that kong-konnect shows as connected.

claude mcp add --transport http kong-konnect https://global.mcp.konghq.com
claude mcp list

VS Code, Cursor and Windsurf

In VS Code, open MCP: Open User Configuration from the command palette and add an entry kong-konnect under servers with "type": "http" and the address. Then start the server via MCP: List Servers and sign in in the browser. Cursor and Windsurf expect an entry under mcpServers with a url and an Authorization header carrying your token.

Step 3: First queries

Start with read-only questions, for example: "List all services in the production control plane along with their routes." The assistant uses search to find the right API operation, get_schema to fetch its parameters and execute to call it. You can also analyse error rates by consumer or status code over the last few hours.

Secure rights and changes

Through execute the assistant can also change routes, plugins or Dev Portal pages. Use a token with as few rights as possible and have the client confirm each write action. For automated workflows, Kong recommends a System Account Access Token tied to a service account with defined permissions.

Browser sign-in to Konnect (in supported clients such as Claude Code and VS Code), a Personal Access Token or a System Account Access Token as a bearer header.

Read: control planes, services, routes, consumers, consumer groups, plugins, vaults, API analytics and Kong documentation. Write: Konnect API operations via execute, starting debug sessions, creating, editing and publishing Dev Portal pages, snippets and Markdown components. What succeeds depends on the rights of the account behind the token.

The server runs at Kong and accesses the Konnect services; according to Kong, regional boundaries and role permissions are respected. Retrieved configurations and analytics data enter the AI client context and therefore reach the model provider. With KAi enabled, the KnowledgeBase tool creates embeddings on AWS Bedrock; according to Kong, no training is performed on that data.

Through execute the assistant can change gateway configuration, such as routes or plugins, and publish Dev Portal content. A token with broad rights makes mistakes costly. Content from Dev Portal pages or request data can carry prompt injection. Use least-privilege tokens and confirm write actions in the client.