The platform MCP server gives an agent the tools of the Agent Hub. With these tools, your AI coding assistant reads what is deployed on the platform, searches the logs, and explains Agent Hub terms, without leaving your development environment. This page explains what the server offers and how you connect your AI coding assistant to it.
What Is the Platform MCP Server?
The platform MCP server is a Model Context Protocol (MCP) server that lets an agent look into a running PAS cluster. Any MCP client can use it, for example your AI coding assistant, or an agent you have developed using PAS Code.
You sign in with your own Agent Hub account, so the server configuration holds no secret.
What Your Assistant Can Do With It
|
Your Assistant Can |
Example Question |
|---|---|
|
Tell you what is deployed and whether it runs |
"Which bundles are deployed on platform pas, and are they all fine?" |
|
Find out why something does not work |
"Why does the invoicing service not start?" |
|
Search the logs |
"Show me the errors of the invoicing service in the last hour." |
|
Show the configuration a deployment runs with |
"Which configuration does the invoicing service use, and where does it come from?" |
|
Tell you which bundles and versions you can deploy |
"Which versions of the invoicing service are available?" |
|
Explain Agent Hub terms |
"What is the difference between a bundle and a component?" |
|
Tell you which LLM providers and virtual API keys a platform has |
"Which virtual API keys does platform dev have, and which agents use them?" |
|
Create a virtual API key for an LLM provider |
"Create a key for the provider OpenAI on dev that permits gpt-4o-mini." |
Permissions and Secrets
-
Every request needs your access token.
-
Deployments, configurations, bundles, LLM providers, and virtual API keys are read with your permissions. Your coding assistant sees and creates only what you may see and create.
-
The deployment and registry tools never answer a secret value of a configuration. A secret value is answered as
[REDACTED]. A reference to a secret in$referencesstays visible, so you can see which secret a configuration uses. -
No tool takes or answers a provider API key. An LLM provider is answered with the last four characters of its key only.
Prerequisites
Before you connect your assistant, make sure that:
-
The PAS CLI is installed, and you are connected to the platform with
pas platform login. -
Your AI coding assistant supports a configuration file in the project, a remote MCP server over streamable HTTP (often named
http), and a fixed OAuth client ID for the server. The platform does not let an assistant register a client of its own.
Connecting Using the Skill
The PAS CLI ships a skill, which connects your assistant to the platform MCP server. Ask your assistant in your own words, for example "Connect Claude Code to the platform" or "Add the platform MCP server". The skill does the following:
-
It selects the platform. When two or more platforms are stored, it asks you which one to use. When no platform is stored, it asks you to run
pas platform login <your platform URL>yourself. -
It reads the address and the client ID with
pas platform mcp info. -
It asks which assistants to configure. You can select more than one.
-
It checks whether the project already has a platform MCP server.
-
It proposes a name for the server, for example
platform-staging. -
It writes the server into the project configuration of each selected assistant, for example .mcp.json for Claude Code.
-
It commits the configuration files in one commit. It does not push.
-
It tells you how to sign in. Refer to Signing In below.
The configuration of the platform MCP server is part of the project. It is committed, so everyone who works on the project shares the same server.
Configure one platform MCP server per project. Every platform MCP server offers the same tools under the same names.
With two servers, your assistant has to tell the platform from the name of the server alone, and a call meant for one platform can change another one.
Connecting Manually
To connect your assistant without the skill, read the values for the active platform:
pas platform mcp info
The output looks like this:
Transport: streamable-http
URL: https://staging-agentgateway.pascode.pas-internal.com/hub/mcp
Client ID: mcp-client
Enter the values into the project configuration of your assistant, under a name for the server that you choose. Every assistant has its own configuration format. E.G., for Claude Code, run the following command from the root of your workspace. It writes .mcp.json:
claude mcp add --transport http --scope project --client-id <client ID> <name> <URL>
Keep the following in mind:
-
Leave out the client ID if
pas platform mcp infoprints noClient IDline. -
Commit the project configuration file, so everyone on the project shares the server.
Signing In
Your assistant opens the identity provider of the platform in your browser, and you sign in there. After the sign-in, the tools of the platform MCP server are available in your assistant.
If the identity provider rejects the sign-in because of the redirect URI, the platform does not accept your assistant yet. The administrator of the platform has to allow the redirect URI of your assistant for the client ID.
Refer to Error Handling Platform MCP Server for further problems and their solutions.
Related Content
Related Documentation: