Skip to main content

Creating Custom Cloud Agents

You can only create and edit AI Agents in the Continue Mission Control web interface. This ensures proper validation, versioning, and team collaboration features.

Creating a Cloud Agent

1

🧭 Navigate to Create Agent

Navigate to the New Agent page.
2

🧩 Fill Out the Agent Form

Configure your Cloud Agent with the following fields:
FieldWhat to EnterExample
NameDisplay name shown in Mission ControlGitHub PR Agent
PromptFirst instruction the agent receivesOpen a GitHub PR to fix the specified issue.
Tools (MCPs)Select built-in or custom MCPsGitHub, PostHog, Supabase, etc.
RulesAdd any organizational rulescontinuedev/gh-pr-commit-workflow
ModelChoose a default LLMClaude Sonnet 4.5
TriggerDetermines when the cloud agent will be invokedGitHub, cron, webhook
Tip: Start with a simple prompt and add complexity through rules and tools. The prompt should be a clear, single instruction that defines the AI Agent’s primary goal.
3

💡 Choose Repository & Automation

In the right panel, you can select the repository where the agent will run, and decide if the agent should automatically kick off on creation.
4

✅ Create and Test

Click “Create Agent” to save and publish it.Your agent is immediately available to run in:
  • Mission Control web interface
  • TUI mode: cn --agent your-org/your-agent-name
  • Headless mode: cn --agent -p your-org/your-agent-name "prompt" --auto

Editing a Cloud Agent

You can edit any AI Agent you own or that has Organization-level access. From the Agents page, click your agent’s name. Change any of the fields in the edit form, then click “Save Changes”.

Example Agent Configurations

Here are proven agent configurations you can create or use as inspiration:

Troubleshooting

Common issues and solutions:
ProblemSolution
Cloud Agent doesn’t complete tasksSimplify the prompt, add more specific instructions, verify tool permissions
Tools aren’t workingCheck MCP configuration, verify API keys and permissions in Hub settings
Inconsistent behaviorAdd rules to enforce consistent patterns, test with various input scenarios
Performance issuesConsider model choice, simplify complex multi-step workflows, optimize tool usage