Kong AI Gateway

Disclaimer

Your use of this download is governed by Stonebranch's Terms of Use.

info

For general information on how integrations work, importing templates, and UAC documentation refer to Overview.

Version Information

Template Name	Extension Name	Version	Status
Kong AI Gateway	ue-kong-ai-gateway	1.0.0	Initial Version.

Refer to Changelog for version history information.

Overview

Kong AI Gateway is a layer in front of large language model providers that presents them through a single OpenAI-compatible interface, applying authentication, routing, and model selection centrally rather than in each client. It also fronts Model Context Protocol (MCP) servers, which expose callable tools that a model can use to gather information or perform work. Both self-managed Kong Gateway deployments and Kong Konnect are supported.

This Universal Task allows customers to send prompts to large language models and run autonomous tool-calling sessions against MCP servers in Kong AI Gateway straight from the Universal Controller. Prompts are supplied inline or from Universal Controller Scripts, responses are returned as plain text or as schema-validated JSON, and token consumption is reported back to the task instance.

Key Features

Feature	Description
Chat Completions	Sends single-turn or multi-turn conversations to any model exposed through Kong and returns the response.
Agentic Tool Calling	Runs an autonomous loop in which the model selects and invokes MCP tools until it produces a final answer.
MCP Tool Discovery	Lists the tools published by one or more MCP servers, with an optional filter restricting which tools a run may use.
Structured Responses	Returns answers as plain text, prompt-described JSON, or JSON validated by the provider against a supplied schema.
Prompt Templating	Resolves prompts held in Universal Controller Scripts, including Universal Controller variable and file references.
Flexible Authentication	Authenticates to Kong with an OAuth2 client-credentials grant or a key-auth API key in a configurable header.

Requirements

This integration requires a Universal Agent and a Python runtime to execute the Universal Task.

Area	Details
Python Version	Requires Python 3.11 or 3.14, tested with Agent bundled python distribution.
Universal Agent Compatibility	Compatible with Universal Agent for Windows x64 and version >= 7.8.0.0. Compatible with Universal Agent for Linux and version >= 7.8.0.0.
Universal Controller Compatibility	Universal Controller Version >= 7.8.0.0.
Network and Connectivity	Network connectivity to Kong AI Gateway is required. Populating the dynamic choice fields additionally requires connectivity to the Kong Admin API (Self-Managed) or the regional Konnect API (Konnect).

Supported Actions

There are 3 Top-Level actions controlled by the Action Field:

• Chat: Sends a prompt to a model through Kong and returns the response.
• Agentic MCP: Lets the model call MCP tools repeatedly until it reaches a final answer.
• List MCP Tools: Reports the tools published by the selected MCP servers.

Action: Chat

Sends a system prompt and a user prompt to the selected model and returns the response, optionally as JSON validated against a supplied schema. The task ends as soon as the model replies.

Action: Agentic MCP

Sends the prompt to the model together with the tool catalog discovered from the selected MCP servers, then executes the tools the model requests and feeds the results back. The loop ends when the model produces a final answer, or when the configured maximum number of turns is reached.

Action: List MCP Tools

Connects to each selected MCP server and reports the tools it publishes, together with their descriptions. No model is contacted, so no tokens are consumed.

Action Output

EXTENSION

The extension output provides the following information:

• exit_code, status_description: General info regarding the task execution.
• invocation.fields: The task configuration used for this execution. For more information refer to the exit codes table.
• result.response: The model's reply for the Chat action.
• result.final_answer: The model's concluding reply for the Agentic MCP action.
• result.metadata: Token usage and the reason the model stopped generating.
• result.complete_conversation: The complete message sequence for the Chat action, when requested through the Extension Output field.
• result.tool_call_transcript: The tool-call transcript for the Agentic MCP action, when requested through the Extension Output field. When the transcript is longer than UE_MAX_OUTPUT_RECORDS, only the first entries are embedded and result.transcript_truncated reports the total and shown entry counts.
• result.servers: For the List MCP Tools action, one entry per MCP connection, holding the mcp_connection route path and the names of its tools.
• result.tool_count: For the List MCP Tools action, the total number of tools discovered.
• result.dry_run, result.rendered_conversation, result.resolved_model, result.estimated_prompt_tokens: Returned instead of a model reply when Dry Run is enabled. For the Agentic MCP action, result.discovered_tools and result.allowed_tools_filter_applied are also returned.
• result.errors: List of errors that might have occurred during execution.

Examples:

Successful execution

{
  "exit_code": 0,
  "status_description": "Task executed successfully",
  "invocation": {
    "extension": "ue-kong-ai-gateway",
    "version": "1.0.0",
    "fields": {
      "action": "Chat",
      "kong_deployment": "Konnect (AI Gateway)",
      "auth_method": "Kong API Key (key-auth)",
      "credential": "Credential(user='kong-client', password='***', token='***', key_location='', passphrase='***')",
      "api_key_header_name": "Authorization",
      "kong_gateway_url": "https://kong.example.com",
      "verify_ssl": true,
      "llm_service": "/",
      "model": "azure-openai",
      "system_prompt_text": "You are a release management assistant.",
      "user_prompt_text": "Summarize the deployment outcome in two sentences.",
      "response_format_type": "Text",
      "stdout_options": "Final Result Only",
      "output_options": ["Token Usage and Metadata"]
    }
  },
  "result": {
    "response": "The deployment completed successfully across all four regions. No rollback was required and all health checks reported green.",
    "metadata": {
      "prompt_tokens": 42,
      "completion_tokens": 27,
      "total_tokens": 69,
      "finish_reason": "stop"
    }
  }
}

Failed execution

{
  "exit_code": 2,
  "status_description": "Authentication Error: Kong rejected the credentials.",
  "invocation": {
    "extension": "ue-kong-ai-gateway",
    "version": "1.0.0",
    "fields": {
      "action": "Chat",
      "kong_deployment": "Konnect (AI Gateway)",
      "auth_method": "Kong API Key (key-auth)",
      "credential": "Credential(user='kong-client', password='***', token='***', key_location='', passphrase='***')",
      "api_key_header_name": "Authorization",
      "kong_gateway_url": "https://kong.example.com",
      "verify_ssl": true,
      "llm_service": "/",
      "model": "azure-openai",
      "user_prompt_text": "Summarize the deployment outcome in two sentences.",
      "response_format_type": "Text"
    }
  },
  "result": {
    "errors": ["Authentication Error: Kong rejected the credentials."]
  }
}

STDOUT

The content written to STDOUT is controlled by the STDOUT field. "Final Result Only" prints the chat response or the agentic final answer on its own. "Full Output" prints the entire conversation for the Chat action, and the live tool-call progress for the Agentic MCP action. The List MCP Tools action prints the discovered tool catalog as a table.

Example:

The deployment completed successfully across all four regions. No rollback was
required and all health checks reported green.

Configuration Examples

Example: Key-auth authentication, inline prompts, text response

A Chat task authenticating with a key-auth API key presented in the Authorization header, where the Bearer scheme is added automatically. Kong Gateway URL points at the Kong proxy, LLM Service holds the route path serving chat traffic, and Model names the model to invoke. Both prompts are typed directly into the form, Response Format is left at "Text", and Extension Output requests "Token Usage and Metadata" so that the token counts are also included in the Extension Output; the Output Only fields are populated either way. The task ends as soon as the model replies.

Example: OAuth2 authentication, agentic tool calling, restricted tool set

An Agentic MCP task authenticating with an OAuth2 client-credentials grant, where the Kong Credential holds the Client ID as the user and the Client Secret as the password. Activate Dynamic Choice Fields lets the LLM Service, Model, MCP Connection(s) and Allowed Tools fields be populated from the Kong Admin API. One or more MCP route paths are selected in MCP Connection(s), and Allowed Tools can restrict the run to an approved subset of the discovered catalog. Max Turns caps how many times the model may call tools before the run is stopped, and Poll Interval (s) paces the turns. Save Options writes the final answer to the file given in Save Destination on the Agent, and Extension Output requests "Tool-Call Transcript" so the sequence of tool invocations is recorded in the Extension Output.

Example: Dry run validation, prompts from Universal Controller Scripts

A Chat task with Dry Run enabled, used to validate a configuration before spending tokens. System Prompt from UAC Script and User Prompt from UAC Script are both enabled, so the prompts are read from Universal Controller Scripts and any variable or file references they contain are resolved. Because Dry Run is enabled, no request reaches the model; the extension returns the fully rendered conversation and an estimated prompt token count so the templating can be confirmed.

Input Fields

Name	Type	Description	Version Information
Action	Choice	The operation performed against Kong AI Gateway. Available options: Chat (default) - a single request and response. Agentic MCP - the model works through a task using MCP tools. List MCP Tools - reports the tools each MCP server offers.	Introduced in 1.0.0
Kong Deployment	Choice	The Kong deployment that serves the configuration read by the discovery dropdowns. Available options: Konnect (AI Gateway) (default) - Models and MCP Servers are read from the Konnect API. Self-Managed (Admin API) - tagged Services and ai-proxy plugins are read from the Kong Admin API.	Introduced in 1.0.0
Authentication Method	Choice	The mechanism used to authenticate to Kong on every proxy call. Available options: OAuth2 (Client Credentials) (default) - a client-credentials grant against the route's own token endpoint. Kong API Key (key-auth) - a static API key presented in a header.	Introduced in 1.0.0
Kong Credential	Credential	The Resolvable Credential holding the Kong credentials. For OAuth2, the user is the Client ID and the password is the Client Secret. For key-auth, the password holds the API key, or the token if the key exceeds 255 characters.	Introduced in 1.0.0
API Key Header Name	Text	The header Kong reads the API key from. Kong's key-auth plugin uses apikey by default. For a bearer-style route, Authorization is used and the Bearer scheme is added automatically. This field is visible when Authentication Method is "Kong API Key (key-auth)".	Introduced in 1.0.0
Kong Gateway URL	Text	The address of the Kong proxy serving chat and MCP traffic, for example https://kong.example.com. This is not the Admin API or Konnect API address. HTTPS is required for OAuth2 unless TLS terminates at a load balancer.	Introduced in 1.0.0
Enable SSL Verification	Checkbox	Check to verify Kong's certificate on every connection. Clear only for a test Kong presenting a self-signed certificate; otherwise supply the certificate path below.	Introduced in 1.0.0
SSL Certificate Path	Text	The full path on the agent to the CA certificate file, required when Kong uses a private or self-signed certificate. Left empty when Kong's certificate comes from a public authority. This field is visible when Enable SSL Verification is "true".	Introduced in 1.0.0
Activate Dynamic Choice Fields	Checkbox	Check to enable discovery for the AI Gateway, LLM Service, Model and MCP Connection(s) dropdowns through the Konnect API or the Kong Admin API. Allowed Tools works without it.	Introduced in 1.0.0
Activate Override Fields	Checkbox	Check to show the free-text Override fields for the multi-select dropdowns, for use when discovery cannot populate them. This field is visible when Action is "Agentic MCP" or "List MCP Tools".	Introduced in 1.0.0
Kong Admin API URL	Text	The API host used for discovery. For Konnect, the regional API host such as https://eu.api.konghq.com, not the *.cp.konghq.com control-plane host. For a self-managed deployment, the Kong Admin API such as http://kong-host:8001. This field is visible when Activate Dynamic Choice Fields is "true".	Introduced in 1.0.0
Kong Admin API Token	Credential	The Resolvable Credential holding the discovery token, which is never used at task runtime. Konnect requires a Personal or System Account token. A self-managed deployment uses an Enterprise RBAC token, and none on Kong OSS. The token is supplied bare, without a Bearer prefix. This field is visible when Activate Dynamic Choice Fields is "true".	Introduced in 1.0.0
AI Gateway	Dynamic Choice	The Konnect AI Gateway holding the Models and MCP Servers, selected from the dropdown or typed as a UUID. The LLM Service, Model and MCP Connection(s) dropdowns require it to be set first. This field is visible when Kong Deployment is "Konnect (AI Gateway)".	Introduced in 1.0.0
LLM Service	Dynamic Choice	The Kong route path serving chat traffic, with an optional leading slash. On Konnect this is usually /. On a self-managed deployment these are the routes tagged llm. A path the dropdown does not list can be typed directly. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
Model	Dynamic Choice	The model to invoke. Required on Konnect, where every model shares one route path and is selected by name. On a self-managed deployment this is left empty so the route's ai-proxy configuration decides. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
MCP Connection(s)	Dynamic Choice	One or more Kong MCP route paths, each with an optional leading slash. On Konnect these come from MCP Server entities. On a self-managed deployment they are the Services and Routes tagged mcp. This field is visible when Action is "Agentic MCP" or "List MCP Tools".	Introduced in 1.0.0
MCP Connection(s) (Override)	Text	A comma-separated list of Kong MCP route paths, for example /mcp/github,/mcp/jira. Overrides the MCP Connection(s) dropdown when populated. This field is visible when Activate Override Fields is "true".	Introduced in 1.0.0
Allowed Tools	Dynamic Choice	The subset of tools the run may use, read from the selected MCP connections. An empty value allows all discovered tools. This field is visible when Action is "Agentic MCP".	Introduced in 1.0.0
Allowed Tools (Override)	Text	A comma-separated list of tool names, where an empty value allows all tools. Overrides the Allowed Tools dropdown when populated. This field is visible when Activate Override Fields is "true".	Introduced in 1.0.0
System Prompt from UAC Script	Checkbox	Check to read the system prompt from a Universal Controller Script instead of the inline System Prompt field. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
System Prompt	Text	The instructions that set the model's role and behaviour, for example "You are a support agent." Left empty when the user prompt is sufficient on its own. This field is visible when System Prompt from UAC Script is "false".	Introduced in 1.0.0
System Prompt Script	Script	The Universal Controller Data Script holding the system prompt, for when it is reused or version-controlled rather than typed into the form. This field is visible when System Prompt from UAC Script is "true".	Introduced in 1.0.0
User Prompt from UAC Script	Checkbox	Check to read the user prompt from a Universal Controller Script instead of the inline User Prompt field. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
User Prompt	Text	The question or task sent to the model. For a multi-turn conversation, [User] or [Assistant] is placed on its own line above each message. This field is visible when User Prompt from UAC Script is "false".	Introduced in 1.0.0
User Prompt Script	Script	The Universal Controller Script holding the prompt or conversation, for when it is reused or version-controlled rather than typed into the form. This field is visible when User Prompt from UAC Script is "true".	Introduced in 1.0.0
Response Format	Choice	The shape of the answer returned by the model. Available options: Text (default) - plain text. JSON (Prompt Described) - JSON described in the prompt itself. JSON Schema (API Enforced) - JSON validated by the provider against a supplied schema. This field is visible when Action is "Chat".	Introduced in 1.0.0
Response Schema	Script	The Universal Controller Script holding the JSON Schema the answer must follow, given either as the schema itself or as an object with schema, name and strict keys. This field is visible when Response Format is "JSON Schema (API Enforced)".	Introduced in 1.0.0
Advanced Options	Checkbox	Check to show the model tuning options covering creativity, answer length and repetition. Left clear, the model's own defaults apply. This field is visible when Action is "Chat".	Introduced in 1.0.0
Temperature	Float	How varied the answers are, from 0.0 to 5.0. Lower values are focused and repeatable, higher values more creative. 1.0 is the usual starting point. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Top P	Float	An alternative to Temperature that keeps the model to the most likely wording. 1.0 considers everything and lower values narrow it. One or the other is adjusted, not both. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Frequency Penalty	Float	How strongly repetition of the same words is discouraged, from -2.0 to 2.0. Values above 0 reduce repetition and values below 0 allow more. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Presence Penalty	Float	How strongly the model is encouraged to introduce new topics, from -2.0 to 2.0. Values above 0 push for variety and values below 0 keep it on topic. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Max Tokens	Integer	The longest answer the model may produce, in tokens, where 1000 tokens is roughly 750 words. The effective ceiling is the model's own limit. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Seed	Integer	A number from 0 to 2147483647 that produces the same answer for the same prompt, where the model supports it. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Stop Sequences	Text	A comma-separated list of sequences that end generation, for example ###, END. This field is visible when Advanced Options is "true".	Introduced in 1.0.0
Max Turns	Integer	The number of times, from 1 to 100, that the model may exchange with the tools before the run is stopped. This field is visible when Action is "Agentic MCP".	Introduced in 1.0.0
Poll Interval (s)	Integer	The number of seconds to wait between tool-calling turns, used to pace the run. 0 sends the next turn immediately. This field is visible when Action is "Agentic MCP".	Introduced in 1.0.0
Dry Run	Checkbox	Check to preview the configuration without calling the model. The Chat action sends nothing at all, while the Agentic MCP action still connects to the MCP servers to list their tools but calls none of them. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
Save Options	Choice	Whether output is persisted to a local file on the agent. Available options: -- None -- (default) - nothing is written to disk. Save Latest Response / Final Answer - only the final reply is written. Save Entire Conversation - the complete message sequence is written. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
Save Destination	Text	The local file path written on the agent, for example /opt/uac/output/kong_result.txt. This field is visible when Save Options is "Save Latest Response / Final Answer" or "Save Entire Conversation".	Introduced in 1.0.0
STDOUT	Choice	What the task writes to STDOUT. Available options: Final Result Only (default) - the chat response or agentic final answer. Full Output - the entire conversation, or live tool-call progress for the Agentic MCP action. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0
Extension Output	Multiple Choice	The optional result sections included in the Extension Output, where the response or final answer is always included. Available options: Token Usage and Metadata (default) - token counts and the reason the model stopped. Complete Conversation - the full message sequence, for the Chat action only. Tool-Call Transcript - the sequence of tool invocations, for the Agentic MCP action only. This field is visible when Action is "Chat" or "Agentic MCP".	Introduced in 1.0.0

Output Fields

The following output-only fields provide better visibility during the execution of Universal Task Instances.

Name	Type	Description	Version Information
Prompt Tokens	Integer	The number of tokens consumed by the prompt, for the Chat action.	Introduced in 1.0.0
Completion Tokens	Integer	The number of tokens consumed by the model's reply, for the Chat action.	Introduced in 1.0.0
Total Tokens	Integer	The combined prompt and completion token count, for the Chat action.	Introduced in 1.0.0
Finish Reason	Text	The reason the model stopped generating, for the Chat action.	Introduced in 1.0.0
Turns Used	Integer	The number of model round-trips consumed, for the Agentic MCP action.	Introduced in 1.0.0
Tool Call Count	Integer	The number of tool invocations performed, for the Agentic MCP action.	Introduced in 1.0.0
Tool Count	Integer	The number of tools discovered, for the List MCP Tools action.	Introduced in 1.0.0

Environment Variables

Environment Variables can be set from the Environment Variables task definition table.

Environment Variable Name	Description	Version Information
UE_HTTP_TIMEOUT	The timeout in seconds applied to every outbound call to Kong. Defaults to 60.	Introduced in 1.0.0
UE_MAX_OUTPUT_RECORDS	The maximum number of transcript entries embedded inline in the Extension Output. Defaults to 100.	Introduced in 1.0.0

Importable Configuration Examples

This integration provides importable configuration examples along with their dependencies, grouped as Use Cases to better describe end to end capabilities.

warning

These examples aid in allowing task authors to get more familiar with the configuration of tasks and related Use Cases. Such tasks should be imported in a Test system and should not be used directly in production.

Initial Preparation Steps

• STEP 1: Go to Stonebranch Integration Hub and download the "Kong AI Gateway" integration. Extract the downloaded archive in a local directory.
• STEP 2: Locate and import the integration to the target Universal Controller. For more information refer to the How To section in this document.
• STEP 3: Inside the directory named "configuration_examples" you will find a list of definition zip files. Upload them one by one in the order given below, by using the "Upload" functionality of Universal Controller:
  • 1_variables.zip
  • 2_credentials.zip
  • 3_tasks.zip
  • 4_workflows.zip
• STEP 4: Update the uploaded UAC Credential "UC1: Kong AI Gateway - Kong API Key Credentials" with your Kong API key in the "Runtime Password" field.
• STEP 5: Update the UAC global variables introduced with the 1_variables.zip file. Their names are prefixed with "ue_kong_ai_gateway". Review the descriptions of the variables as they include information on how they should be populated.

tip

• The order indicated above ensures that the dependencies of the imported entities need to be uploaded first.
• All imported entities are prefixed with UC1: Kong AI Gateway.

How to "Upload" Definition Files to a Universal Controller

The "Upload" functionality of Universal Controller allows Users to import definitions exported with the "Download" functionality.

Login to Universal Controller and:

• STEP 1: Click "Tasks" → "All Tasks"
• STEP 2: Right click on the top of the column named "Name"
• STEP 3: Click "Upload..."

In the pop-up "Upload..." dialogue:

• STEP 1: Click "Choose File".
• STEP 2: Select the appropriate zip definition file and click "Upload".
• STEP 3: Observe the Console for possible errors.

Use Case 1: AI Briefing

Description

This workflow asks an AI Agent a question and lets it answer using the MCP tools published through Kong AI Gateway. The first task lists the available MCP tools, which confirms that the gateway and its MCP servers are reachable without consuming any LLM tokens. Only when that succeeds does the second task run the AI Agent, which calls those tools to gather facts, answers the question, and saves the answer as a Markdown file on the Agent.

The question, the Kong routes, the model and the report location are all UAC variables, so the same workflow can be scheduled as a recurring briefing or launched on demand with a different question.

The tasks configured demonstrate the following capabilities among others:

• Inventory the MCP tools available through Kong without a model call.
• Let the model decide which MCP tools to call, capped by a turn limit.
• Save the AI Agent's answer to a file on the Agent for downstream use.
• Report token usage on every task instance.

The components of the solution are described below:

1. "UC1: Kong AI Gateway - List Available MCP Tools" - Lists the tools published through the MCP connection in ue_kong_ai_gateway_mcp_connection. No model is contacted, so no tokens are consumed.
2. "UC1: Kong AI Gateway - Ask the AI Agent" - Runs only if (1) succeeds. Sends the question in ue_kong_ai_gateway_question to the model with those tools, for up to 5 turns, prints each tool call to STDOUT, and saves the final answer to ue_kong_ai_gateway_report_path.

How to Run

Execution Steps:

• STEP 1: Ensure all Initial Preparation Steps have been completed successfully.
• STEP 2: Launch the workflow "UC1: Kong AI Gateway - AI Briefing".
• STEP 3: Review the STDOUT and the Output Only fields of (1) and (2), and the file at the configured report path.

Expected Results:

• Task (1) lists the discovered MCP tools and populates the Tool Count field.
• Task (2) prints the tool calls it made and their results, and writes the final answer to the report path on the Agent.
• The token counts, turns used and tool-call count appear in the Output Only fields of task (2).
• The workflow completes in the Success state.

Exit Codes

Exit Code	Status	Description
0	Success	Successful Execution.
1	Failure	Generic Error. Raised when not falling into the other Error Codes.
2	Failure	Authentication error. The provided credentials are invalid or insufficient.
20	Failure	Input fields validation error.

Observability

Token usage metrics are generated and exported by this integration, following the OpenTelemetry semantic conventions for generative AI. They are emitted only when the Universal Agent is OpenTelemetry-compatible and metrics are enabled for the task.

Metric: gen_ai.client.token.usage

Name	Instrument Type	Unit (UCUM)	Attributes	Description
gen_ai.client.token.usage	Histogram	{token}	As defined on Metric Attributes List	Number of input and output tokens used, recorded once per model call.

Metric Attributes List:

Attribute Name	Description
gen_ai.token.type	input for prompt tokens, output for completion tokens.
gen_ai.operation.name	The operation performed, chat.
gen_ai.provider.name	The API dialect used towards Kong, openai.
gen_ai.request.model	The model requested in the Model field. Omitted when the field is empty.
gen_ai.response.model	The model that produced the response, as reported by Kong.
server.address	The host name of the Kong Gateway URL.
server.port	The port of the Kong Gateway URL.

Document References

Document Link	Description
Kong AI Gateway Documentation	Covers how Kong AI Gateway exposes model providers through a single interface, and how models and routes are configured.
Kong AI Proxy Plugin Reference	Describes the plugin that maps a Kong route onto an upstream model provider, which determines the route paths and model names entered in this task.
Kong Key Authentication Plugin Reference	Documents key-auth configuration, including the header names the key can be presented in.
Kong OAuth 2.0 Authentication Plugin Reference	Documents the OAuth2 plugin and the client-credentials grant used to obtain a token from the route's own token endpoint.
Kong Model Context Protocol Documentation	Explains how MCP servers are published through Kong, which determines the MCP route paths and the tools available to an agentic run.

Changelog

ue-kong-ai-gateway-1.0.0 (2026-10-20)

Initial Version