Configure coding agents centrally

Use a coding agent configuration to give your team a shared setup for agents, models, MCP servers, and skills. Publish the configuration for developers in your Databricks workspace, and the Unity Gateway CLI (ug) fetches it and applies it on each developer's machine.

Beta

Coding agent configuration is in Beta. An account admin must enable the Enhanced Unity Gateway preview from the account console Previews page. See Manage Databricks previews.

Coding agent configurations

A workspace admin publishes a coding agent configuration to a Databricks workspace. Developers apply it to their machines with ug configure.

The configuration defines:

Setting	What it configures
Coding agents	The enabled agents and the default agent.
Models	Each agent's model source and default models. Use Databricks models or an external provider registered in Unity Catalog.
MCP servers	The servers to connect to the enabled agents.
Skills	The shared skills to make available to the enabled agents.
Unity Gateway features	Agent settings such as Smart Routing and tracing, where supported.

Setting	What it configures
Coding agents	The enabled agents and the default agent.
Models	Each agent's model source and default models. Use Databricks models or an external provider registered in Unity Catalog.
MCP servers	The servers to connect to the enabled agents.
Skills	The shared skills to make available to the enabled agents.
Unity Gateway features	Agent settings such as Smart Routing and tracing, where supported.

Reference models, MCP servers, and skills by their Unity Catalog names, or select them from a catalog and schema. For external models, see Bring your own model providers.

Prepare resources

Before you publish, set up the models your team needs and grant developers access to them, along with any MCP servers and skills you include. See Set up model capacity.

Publish a configuration

Use the UI

As a workspace admin:

1. Open Unity Gateway > Govern > Coding Agents.
2. Click Edit configuration to open the JSON editor.
3. Update the agents, models, MCP servers, skills, and features in the configuration. Click Format JSON to format the JSON.
4. Click Save to publish the configuration for developers in this workspace.

Choose model sources for each agent

For each enabled agent, set exactly one model source in config.models. Agents in the same configuration can use different model source types.

1. Static list of model services (recommended): Set model_services to the full Unity Catalog names of the model services to make available. Every model service must support an API that is compatible with the agent. For example, Claude Code requires the Anthropic Messages API. Codex requires the Open Responses API or OpenAI Responses API.

   JSON

   {
     "models": {
       "model_services": ["system.ai.claude-sonnet-4-6", "system.ai.claude-opus-4-8"]
     }
   }

For Smart Routing model-source requirements, see Smart Routing limitations.

1. Model provider service: Set model_provider_service to the full Unity Catalog name of a model provider service. All models configured for that model provider service are available to developers.

   JSON

   {
     "models": {
       "model_provider_service": "main.default.anthropic_prod"
     }
   }

2. Unity Catalog location: Set unity_catalog_location to a catalog and schema. All model services in that location for which a developer has EXECUTE permission are available to that developer.

   JSON

   {
     "models": {
       "unity_catalog_location": "system.ai"
     }
   }

Configuration example

This example configures Claude Code with a default model, one MCP server, and one skill. Replace the resource names with resources your team can access. When editing an existing configuration, preserve any settings your team still needs.

JSON

{
  "default_agent": "CODING_AGENT_CLAUDE_CODE",
  "enabled_agents": [
    {
      "agent": "CODING_AGENT_CLAUDE_CODE",
      "config": {
        "models": {
          "model_services": ["system.ai.claude-sonnet-4-6"]
        },
        "default_models": {
          "default_model": "system.ai.claude-sonnet-4-6"
        }
      }
    }
  ],
  "mcp_servers": { "names": ["main.developer_tools.github"] },
  "skills": { "names": ["main.team_skills.code_review"] }
}

The default_agent must also be listed in enabled_agents.

For details about the overall default and Claude Code family defaults, see Set a default model.

Use the API

To publish from a script, save the JSON as coding-agent-config.json. With the Databricks CLI authenticated to the workspace as a workspace admin, create the configuration:

Bash

databricks api post /api/ai-gateway/v2/coding-agent-configs \
  --json @coding-agent-config.json

The workspace supports one published configuration. To change it, update the existing configuration with PATCH /api/ai-gateway/v2/coding-agent-configs/{id}, using the returned resource name and an update_mask query parameter that identifies the fields to change. Keep the JSON file in version control to review changes before publishing them.

Sync configuration updates

When developers launch an agent through ug, it checks the published configuration and applies updated agent and model settings. Changes can take a few minutes to appear.

To refresh immediately and sync managed MCP servers and skills, run:

Bash

ug configure

Next steps

• Set up model capacity