TachiBot MCP Documentation

What is TachiBot?

TachiBot is a Universal AI Orchestrator that coordinates multiple AI models (GPT-5, Gemini, Perplexity, Grok, Qwen, and more) into collaborative workflows. Instead of relying on a single model, you orchestrate collective intelligence—models brainstorm together, challenge assumptions, cross-verify facts, and combine insights.

Customizability by Design

Instead of fixed tools, TachiBot provides composable building blocks so you can architect exactly the AI behavior you need. You're not locked into predefined workflows—you define your own.

Build AI Flows, No Boilerplate

Custom YAML/JSON workflows let you:

• Chain multiple AI models into complex reasoning pipelines
• Pass outputs between steps with variable interpolation (${step_name.output})
• Control costs with per-workflow budgets and smart model routing
• Optimize automatically with built-in caching and token compression
• Mix and match models - use the best AI for each specific task

Where to Define Workflows

TachiBot auto-discovers workflows from:

• workflows/ - Recommended default (plain folder)
• .tachi/workflows/ - Alternative hidden config location

Create your workflow files in either location and TachiBot instantly recognizes them. No boilerplate, no complex setup.

Core Concept: Collective Intelligence

Like Tachikomas from Ghost in the Shell sharing knowledge through synchronized thinking, TachiBot enables AI models to work as a swarm:

• Diverse Perspectives - Each model brings different strengths (creative, logical, research-backed, technical)
• Cross-Verification - Models verify each other's outputs to eliminate hallucinations
• Debate & Refinement - Make models debate for 5, 10, or even 200 rounds to refine solutions
• Synchronized Synthesis - Combine all insights into actionable, high-confidence recommendations

Key Features

Configurable Tools (400 tokens → 13k tokens)

Enable only the tools you need via tools.config.json. Each tool adds ~400-500 tokens, so you control your context usage:

• Just think tool? ~400 tokens
• All 31 tools enabled? ~13k tokens
• Pick exactly what you need and adjust anytime

Multi-Model Workflows

Chain unlimited steps in YAML/JSON. Example: Brainstorm with GPT-5 → Research with Perplexity → Verify with Grok → Challenge with Challenger → Synthesize final answer.

PingPong Debates

Make AI models debate in 4 different styles (collaborative/competitive/debate/build-upon) for up to 200 rounds. Watch them refine ideas through argumentative discourse.

Proven Performance Benefits

Based on Anthropic research, the think tool alone delivers:

• 54% improvement in complex reasoning tasks
• Better adherence to detailed policies and guidelines
• Enhanced consistency across multiple trials

TachiBot extends this with multi-model orchestration for even better results.

How It Works

TachiBot provides 31 MCP tools you can mix and match:

• Core Tools - think, focus, nextThought (no API keys needed)
• AI Providers - Perplexity (3), Grok (6), OpenAI/GPT-5 (5), Gemini (3), OpenRouter (2-3)
• Workflow Tools - workflow, list_workflows, create_workflow, visualize_workflow, workflow_start, continue_workflow, workflow_status (7 tools)
• Workflow Validators - validate_workflow, validate_workflow_file (2 tools)
• YAML Workflows - verifier, scout, challenger, pingpong (not MCP tools, but workflow files)

You define workflows that chain these tools together. Each step can pass outputs to the next, run in parallel, or branch conditionally.

Example: Swarm Intelligence

The swarm-think workflow demonstrates TachiBot's power:

1. Diverse Sensing - 4 models gather initial perspectives (Gemini creative, GPT-5 logical, Perplexity research, Qwen technical)
2. Synchronization - Extract common patterns and unique insights
3. Real-time Data - Grok searches for latest information
4. Deep Processing - Focus tool coordinates 3 models in collaborative reasoning (5 rounds)
5. Adversarial Challenge - 3 models debate the analysis (3 rounds)
6. Final Synthesis - Combine all perspectives with confidence levels