Supported environment

Before installing Obeo AI for Capella, check the following requirements. Installation steps are provided in the Installation Guide.

1. Software requirements

Component Requirement Notes

Operating system

Windows, Linux or macOS

Obeo AI for Capella can be installed and run in the corresponding Capella distribution on each of these operating systems. Use the Windows archive or the Linux/macOS archive, as appropriate. Capella and Java must match the operating system and processor.

Capella

7.1.0 (target version), 7.0.1 or 6.1.0

The installer checks the underlying Capella version, including for SMW. Capella or SMW must already start successfully with its own runtime. The bundled ELK layout integration is installed for 7.1.0 and 7.0.1; it is skipped for 6.1.0.

MCP and CLI runtime

Java 21 JRE or JDK

Java 21 is required for the MCP and CLI tools with every supported Capella version. With standard Capella 7.1.0, the launchers automatically reuse its bundled Java 21 unless another runtime is configured. For Capella 7.0.1, 6.1.0 or SMW without a bundled Java 21, configure a separate runtime. This does not change the runtime used by Capella or SMW.

Windows installer tools

Command Prompt and PowerShell with Expand-Archive

Required to run the packaged .bat installer and extract the agent archive.

Linux/macOS installer tools

Bash and unzip

Required to run the packaged .sh installer and extract the agent archive.

AI client

An agentic AI tool that can configure and invoke a local MCP server

Use a CLI, desktop application or IDE integration supporting the selected transport: a local command for stdio, or a server URL for HTTP/HTTPS. An LLM account alone is insufficient. See the client combinations tried and LLM client configuration.

For runtime selection and verification, follow Java 21 for the MCP and CLI tools.

2. Permissions on the target host

  • The account performing installation needs write access to the Capella installation and the tools directory to install the features and extract the agent.

  • The account running the product must be allowed to execute Java and the MCP launchers. On Linux and macOS, launcher scripts need executable permission; the installer can be run through Bash as shown in the installation guide.

  • Run the local MCP server under the same user account as Capella so it can access the per-user runtime registry and discover the running instance.

  • The account running Capella needs access to its workspace and models, including write access to save model and session changes.

For obtaining the distribution and a license, see Get Obeo AI for Capella. Follow the Installation Guide for installation steps, and Transport and access rights for stdio/HTTP process permissions and client approvals.

3. Connections and external services

The MCP server runs on the same computer as Capella. Its connection to Capella uses a dynamically assigned loopback port; no fixed Capella server port needs to be configured. Capella asks you to approve local automation connections by default.

  • stdio: the AI client starts the MCP server and communicates through standard input/output. No MCP listening port is needed.

  • Local HTTP: allow the client to reach the configured loopback endpoint. The installation examples use http://127.0.0.1:3333/mcp; the server process must remain running while the client uses it.

  • Remote HTTPS: provide a reachable endpoint, a TLS certificate trusted by the client and bearer authentication. A hosted client cannot access your computer’s loopback address. Follow HTTPS with bearer authentication for setup.

For the account permissions and client approvals needed by each mode, see Transport and access rights.

The AI client also needs access to its configured LLM service and any credentials required by that service. Configure network and proxy access in the client as required by your organization. Installing the bundled update site does not provide an LLM service or account.