Tool Reference

Quick reference for all bab MCP tools. See individual tool docs in tools/ for detailed usage, examples, and best practices.

Utility

  • list_models: Lists available models from configured AI providers and delegate plugins. No parameters.
  • version: Returns bab and runtime version information. No parameters.

Delegate

  • delegate: Runs a prompt through a configured CLI plugin adapter. Bridges bab to external CLIs (Claude Code, Codex, Copilot, OpenCode).

Simple Tools

  • chat: General-purpose collaborative thinking partner for brainstorming, development discussion, second opinions, and exploring ideas.

Workflow Tools (Multi-Step Investigation + Expert Validation)

  • thinkdeep: Multi-stage investigation and deep reasoning for complex problems, architecture decisions, and performance challenges.
  • debug: Systematic debugging with hypothesis testing, evidence collection, and root cause analysis.
  • analyze: Comprehensive code analysis for architecture, performance, security, quality, and pattern understanding.
  • codereview: Systematic code review covering quality, security, performance, and architecture.
  • planner: Interactive task planning with branching, revision, and incremental refinement.
  • refactor: Code smell detection, decomposition planning, modernization, and organization analysis.
  • secaudit: Security audit with OWASP Top 10, compliance evaluation, and threat modeling.
  • testgen: Test suite generation with edge case coverage and failure mode analysis.
  • docgen: Code documentation generation with complexity analysis and call flow mapping.
  • precommit: Pre-commit validation for git changes with security review and impact assessment.
  • tracer: Code tracing in precision mode (execution flow) or dependencies mode (structural relationships).

Specialized Tools

  • consensus: Multi-model consensus through structured debate. Consults multiple AI models (SDK or plugin via pluginId/modelName) with different stances, per-model temperature and thinking mode, optional parallel execution, and a synthesized recommendation.
  • challenge: Forces critical thinking by scrutinizing claims. No AI model call — pure prompt-based.

Common Parameters

Most workflow tools share these parameters:

Parameter Type Description
step string Current investigation step content
step_number number Current step number (>= 1)
total_steps number Estimated total steps (>= 1)
next_step_required boolean Whether more steps are needed
findings string Discoveries and evidence collected
confidence enum exploring, low, medium, high, very_high, almost_certain, certain
continuation_id string Resume previous conversation thread
model string Specific model ID or alias
temperature number Response creativity (0-1)
thinking_mode enum minimal, low, medium, high, max
files_checked string[] All files examined during investigation
relevant_files string[] Files directly relevant to the task
relevant_context string[] Methods/functions involved
issues_found object[] Issues with description and severity
images string[] Absolute paths to screenshots/diagrams
use_assistant_model boolean Whether to run expert validation phase

Conversation Threading

  • Max 8 turns of history per thread
  • Max 16 total turns before requiring a new continuation
  • Use continuation_id to maintain context across multiple tool calls

File Embedding

  • Max file size: 50,000 tokens per file
  • Max total: 40% of model context window (min 4,000, max 50,000 tokens)
  • Must be absolute paths
  • Non-files are skipped with reason

Table of contents