Loading ToolBench
A multi-agent orchestration and task automation platform with built-in tools for filesystem operations, web browsing, memory management, media handling, scheduling, and inter-agent delegation.
GoClaw provides 43 tools with descriptions and basic input schemas, but has significant gaps in schema completeness, parameter descriptions, and error handling guidance. Transport is unknown (not STDIO, not explicit HTTP), which is a critical blocker. Naming is mostly strong (verb-noun patterns), but many parameters lack type information and comprehensive constraints. Output schemas are not documented. Error handling exists but doesn't guide recovery. This is a typical community-grade MCP server with functional tools but production-quality gaps.
Automate browser interactions: navigate pages, click elements, fill forms, take screenshots
Run Claude Code CLI on a remote workstation via workstation_exec
Generate music or sound effects from text descriptions using AI
Create a Telegram forum topic and return its message_thread_id for routing
Generate images from text prompts using an image generation provider
Generate videos from text descriptions using AI
Transport mechanism unknown, cannot verify remote accessibility. Source code does not explicitly declare HTTP, SSE, or STDIO transport.
Output schemas not documented. Tools like exec, browser, cron, heartbeat, spawn, and delegate accept generic action/command parameters but do not document what they return. LLMs cannot plan downstream calls without knowing the response structure.
Parameters with overloaded semantics. 'action' and 'command' parameters in browser, cron, heartbeat accept free-form strings without enum constraints or documented valid values. LLMs will hallucinate invalid actions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Schedule or manage recurring tasks using cron expressions, at-times, or intervals
Get the current date and time with timezone support, for precise timestamps in scheduling and memory
Delegate a task to a linked agent for inter-agent orchestration
Apply targeted search-and-replace edits to existing files without rewriting the entire file
Execute a shell command in the workspace and return stdout/stderr
Schedule or manage the agent's recurring self-check-in heartbeat
Search entities, relationships, and observations in the agent's knowledge graph
List files and directories in a given path within the workspace
List the members of the current group chat
Expand a memory search result with surrounding context from the same document
Retrieve a specific memory document by its file path
Search through the agent's long-term memory using semantic similarity
Send a proactive message to a user on a connected channel (Telegram, Discord, etc.)
Register a skill directory (created via skill-creator) in the system database, making it discoverable and grantable to agents
Analyze audio files (speech, music, sounds) using an audio-capable LLM provider
Analyze documents (PDF, Word, Excel, PowerPoint, CSV, etc.) using a document-capable LLM provider
Read the contents of a file from the agent's workspace by path
Analyze images using a vision-capable LLM provider
Analyze video files using a video-capable LLM provider
Send an existing workspace file as an attachment in the current chat (does not create or modify the file)
Get the current status and metadata of a specific chat session
Retrieve the message history of a specific chat session
List active chat sessions across all channels
Send a message to an active chat session on behalf of the agent
Search for available skills by keyword or description to find relevant capabilities
Spawn a subagent to handle a task in the background
Transcribe voice/audio messages to text using ElevenLabs Scribe or a proxy service
Convert text to natural-sounding speech audio
Activate a skill to use its specialized capabilities (tracing marker)
Read the full content of a Knowledge Vault document by doc_id
Search the Knowledge Vault (documents, wikilinks, episodic and knowledge graph fan-out)
Pause the current agent tool sequence for a bounded number of milliseconds
Fetch a web page or API endpoint and extract its text content
Search the web for information using a search engine (Brave or DuckDuckGo)
Execute an allowlisted command on a linked remote workstation
Write content to a file in the workspace, creating directories as needed
Resolve a Zalo group's real chat ID from its display name
No pagination guidance. Tools like sessions_list, knowledge_graph_search, vault_search, and list_group_members accept a query but do not describe limit, offset, or pagination token parameters. Large result sets will blow context windows.
Error handling does not guide recovery. Tools like exec, web_fetch, and browser will fail silently or return raw error codes without recovery hints (e.g., 'Command not found. Try listing available workstations first.').
Irreversible operations (exec, write_file, edit, delete, send, create_forum_topic, cron, spawn, delegate) lack confirmation or dry-run steps. Agents can accidentally run destructive commands or send messages.
Tool descriptions lack context on when to use them. Similar tools (e.g., memory_search vs vault_search vs knowledge_graph_search) have overlapping purposes but no guidance on which to call for a given use case.
No mention of permission checks or scope declarations. Tools like write_file, exec, sessions_send, and delegate can perform sensitive actions but lack guidance on required permissions or authorization.
Parameter descriptions missing for generic inputs. 'action' in cron and heartbeat, 'command' in workstation_exec, and 'prompt' in create_image/create_video/create_audio lack format guidance or valid value lists.
No multi-step composition guidance. Tools like spawn and delegate accept task descriptions but do not document expected input format, return values, or how to compose results into the agent's plan.