Find answers to common questions about Boundary. If you don't find what you're looking for, check our documentation or contact support.
Boundary runs as a Docker container, so you need Docker Desktop or Docker Engine installed. You also need Cursor IDE for the MCP integration, and at least one LLM API key (OpenAI, Anthropic, or Google).
No, you only need at least one. Boundary will use OpenAI by default if available, but you can configure it to use any of the three providers. Having multiple keys allows you to use different providers for different agents.
Boundary is designed to work with Cursor through the Model Context Protocol (MCP). Cursor provides the code intelligence that Boundary needs to understand your codebase. While technically possible to use Boundary with other MCP-compatible tools, Cursor is the recommended and best-supported option.
After setting up the MCP connection in Cursor, you can verify it's working by asking Cursor to list available MCP tools. You should see Boundary's debate tools (debate_start, debate_continue, record_decision) available. You can also try initiating a simple debate to test the connection.
Good debate questions are specific, focused, and clearly state what you're deciding. For example, 'Should we add caching to AuthService?' is better than 'Should we optimize our code?' The more specific your question, the better Boundary can generate targeted queries and provide relevant analysis.
Debate duration depends on the complexity of your question and codebase. Simple decisions may complete in a few minutes, while complex architectural decisions may take 10-15 minutes as Cursor and Boundary work through multiple rounds of context gathering. The iterative process ensures comprehensive analysis.
Providing relevant context files can help, but it's not required. Boundary will generate semantic queries about your codebase, and Cursor will investigate using its code intelligence. However, if you know specific files or directories are particularly relevant, including them can speed up the process.
Decision maps don't provide a single answer. They illuminate risk surfaces, trade-offs, and failure modes. Focus on areas where agents disagree (these indicate significant trade-offs), irreversible commitments (decisions that will be hard to undo), and specific risk surfaces (concrete failure vectors). See our Decision Maps guide for detailed guidance.
Record decisions after you've reviewed the decision map and made your final choice. Recording creates a permanent, queryable record that helps inform future decisions when similar boundaries emerge. It's especially valuable for significant architectural decisions.
The Free tier includes limited debates per installation, access to all core agent types, and local decision storage. It does not include decision history persistence, export capabilities, web dashboard access, cloud sync, or team collaboration features.
Individual is €9/month for 1 seat with Cloud Sync and web dashboard. Team is €49/month for the first 10 seats, plus €5/seat/month beyond 10, with the same cloud features plus team collaboration and org-wide search. Both include unlimited debates, decision history, and export.
The Lifetime tier is a one-time payment of €499 that includes full access to all features, unlimited seats, unlimited debate sessions, all agent types, and optional cloud sync and web dashboard access. It's fully on-premise by default, with cloud features as optional add-ons.
Yes, you can change your subscription tier at any time. Upgrades take effect immediately, while downgrades typically take effect at the end of your current billing period. Contact support for assistance with tier changes.
Boundary is cloud-first (Cloud Sync + dashboard). The MCP is open source and you can fork and run it locally so data stays on your infrastructure when you need that. When run locally or self-hosted, decision reports and debate data are stored on your machine or infrastructure and do not leave your environment unless you use Cloud Sync.
Boundary doesn't directly access your codebase. Instead, it works through Cursor's code intelligence. When you initiate a debate, Boundary generates semantic queries about your code. Cursor then uses LSP, 'Go to Definition', 'Find References', and codebase search to gather context and provide it back to Boundary.
Yes, you can customize agent participation through the debate_start tool parameters. You can specify which agent types to include, add custom agents with specific biases, and control the number of debate rounds.
Boundary works with OpenAI, Anthropic, and Google LLM providers. The specific models used depend on your API key configuration. By default, Boundary uses OpenAI if available, but you can configure it to use any of the three providers or mix them across different agents.
Boundary requires internet connectivity to access LLM APIs (OpenAI, Anthropic, or Google). However, once debates are complete and decision maps are generated, you can view and use them offline. When run locally or self-hosted, all stored data is available offline.
First, verify that Docker is running and the Boundary container can start. Check that your MCP configuration in Cursor is correct, including the Docker command and environment variables. Ensure your API keys are set correctly. Try restarting Cursor after making configuration changes.
Long debates are normal for complex decisions, but if you're experiencing timeouts, try: providing more specific context files, simplifying your question, checking your API key rate limits, or reducing the number of debate rounds. Also ensure your internet connection is stable.
Verify that your API keys are set correctly in your environment variables or Cursor's MCP configuration. Check that the keys are valid and have sufficient credits/quota. Ensure the keys have the necessary permissions for the LLM provider you're using.
Incomplete decision maps often indicate that Boundary didn't receive enough context. Try: providing more specific context files, letting the debate run through more rounds, ensuring Cursor can access your codebase, or rephrasing your question to be more specific.
Check the documentation first, especially the Getting Started and Using with Cursor guides. If you're still stuck, contact support at [email protected].
If you can't find the answer you're looking for, we're here to help: