fix: expand session MCP wiring + scheduling infinite loop

[rudiheydra]

Summary

• fix: wire MCP server into ExpandChatSession for feature creation — The expand session (Web UI "Add Features" flow) was silently failing to create features because it had no MCP servers configured and only allowed Read/Glob tools. The prompt told Claude to call feature_create_bulk but the tool was unavailable. This wires the features MCP server into ExpandChatSession (matching how AssistantChatSession already works) and removes the duplicated direct-DB _create_features_bulk() method and <features_to_create> regex parsing.

• fix: prevent infinite loop in compute_scheduling_scores with circular deps — parallel_orchestrator.py could hang when computing scheduling scores if features had circular dependencies. Adds a visited set to break cycles.

Test plan

• Start the UI (python start_ui.py), select a project with 100% features, click "Add Features"
• Chat with Claude about new features, approve the list
• Verify feature_create_bulk MCP tool is called and features appear on kanban board
• Create features with circular dependencies, verify orchestrator does not hang

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Resolved infinite loop issue in the dependency scheduler when circular dependencies are present, preventing scheduling calculations from stalling and improving system stability.

• Improvements

  • Enhanced feature creation workflow with updated system integration for more robust and efficient feature management operations.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR adds circular dependency handling to the BFS-based depth calculation and replaces the REST-based feature creation system with an MCP-based workflow. The first change prevents infinite loops from circular dependencies, while the second refactors feature creation to use MCP server tools instead of in-file parsing logic.

Changes

Cohort / File(s)	Summary
Bug Fix: Circular Dependency Prevention api/dependency_resolver.py	Added visited set to compute_scheduling_scores BFS algorithm to prevent infinite loops from circular dependencies. Nodes are now processed once with deduplication of enqueueing children.
Refactor: MCP-Based Feature Creation server/services/expand_chat_session.py	Replaced REST-based feature creation with MCP-based workflow (feature_create_bulk). Removed in-file feature parsing logic, _create_features_bulk method, and regex parsing. Added EXPAND_FEATURE_TOOLS constant and MCP server configuration to Claude client initialization with bypassPermissions handling. Import changes: removed re, added sys.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐰 A cycle was caught in our graph's gentle dance,
A visited set gives each node its chance.
REST gave way to MCP's bright tools,
Features now flow through smarter rules!
The chat expands with a cleaner way,
Round and round—but no more delay! 🔄

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes both main changes: MCP wiring fix for expand_chat_session and the infinite loop fix in compute_scheduling_scores for circular dependencies.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[leonvanzyl]

Thanks!