Loading ToolBench
TachiBot MCP: 67 AI tools, 12 providers. Multi-model orchestration (Perplexity, Grok, OpenAI, Gemini, Qwen, Kimi, MiniMax, DeepSeek, GLM, StepFun, ERNIE, local), YAML workflows, token-optimized profiles. Smart routing, parallel execution, jury system.
TachiBot MCP presents 13 tools with moderate schema coverage but significant gaps in descriptions and parameter clarity. Schema definitions are visible and typed, but many parameter descriptions are vague or missing critical details. The focus tool family (7/13 tools) operates on session management with reasonable structure, but lacks actionable error guidance. Tools like blog_writer, debug_triage, and diff_review have complex multi-parameter inputs with incomplete constraint documentation. Output schemas are not documented for any tool. Parameter descriptions often fail to explain when/why to use them or what format/constraints apply. No tools declare permissions, no tools use enums to constrain free-form string inputs where appropriate (e.g., focus mode names), and error handling is absent from the visible code. The server claims 67 AI tools but only 13 are shown here; cannot evaluate the full surface.
Write a researched long-form post: interview four personas for the angle, then plan/draft/restructure/line-edit in separate passes. Put the SUBJECT in the 'topic' parameter.
Systematic bug triage (Grok 4.3): RANKED root-cause hypotheses with likelihoods, the cheapest discriminating check for each, and the minimal fix for the top candidate. Provide the error/stack trace in 'error'.
Multi-model diff-aware code review: 2-3 lab-diverse reviewers scoped to the changed lines, deduplicated and severity-ranked by a judge into a merge verdict. Provide the unified diff in 'diff'.
Diagnose your TachiBot setup: which API keys are detected, which tools are available vs hidden (and why), the active profile, and a suggested first step. Zero-cost, needs no API key. Call it when tools seem missing.
Multi-model reasoning with various modes
No output schemas documented for any tool. LLMs cannot predict what fields to expect, forcing trial-and-error or mid-chain discovery calls.
focus tool: mode parameter accepts free-form strings but should be an enum. Visible modes include 'simple', 'focus-deep', 'deep-reasoning', 'code-brainstorm', 'dynamic-debate', etc., but no enum constraint in schema. LLMs will hallucinate invalid modes.
Parameter descriptions lack actionable constraints. 'modes' array in focus tool lists examples in description (vague), not schema enum. 'rounds' and 'temperature' lack min/max bounds. 'maxTokensPerRound' has no guidance on practical ranges.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 55 | - | v1 |
Analyze multiple sessions for patterns, model performance, and insights
Delete old sessions to free up disk space
Export a session in different formats for sharing or archiving
Find sessions similar to a given query for learning from past interactions
List all saved Focus MCP brainstorming sessions with metadata
Get recommendations based on a query and past session patterns
Replay a saved session to see all model responses and synthesis
Get current session management statistics
focus tool: domain parameter lacks enum constraint. Description lists allowed values (ARCHITECTURE, SECURITY, PERFORMANCE, ALGORITHM, CODE, DATA_STRUCTURES) but schema does not enforce them. Should be an enum.
No error recovery guidance documented. Tools like focus_clear_old_sessions (destructive) have no confirmation/dry-run variant. debug_triage and diff_review could fail in many ways but provide no guidance on recovery.
doctor tool has empty input schema ({}). Its description says 'Call it when tools seem missing' but provides no guidance on what output to expect or how to interpret results. Schema is missing entirely.
blog_writer description mentions required 'topic' parameter in all caps but schema does not mark it as required. Schema validation rules are not visible.
focus_clear_old_sessions is a destructive operation (DESTRUCTIVE risk tag) but has no dry-run mode, confirmation step, or warning in the description. Pattern: confirmation-request is unimplemented.
debug_triage and diff_review accept 'files' parameters (file paths or ranges like 'src/foo.ts:100-200') but no validation guidance is documented. Risk of path traversal or reading sensitive files is not addressed.
No tool declares required permissions (read:session, write:session, delete:session, etc.). Cannot assess least-privilege or audit control.