Loading ToolBench
MCP server that exposes PraisonAI tools for Claude Desktop, Cursor, and other MCP clients
PraisonAI MCP exposes 10 tools with basic descriptions and schemas, but falls short of production quality. Tool names lack clear action verbs (run_agent, run_research, run_auto_agents are vague about what 'run' means). Parameter descriptions are minimal (10-30 chars, well below the 72-char baseline). Output schemas are not documented, callers cannot predict response structure. Error handling returns generic dicts without recovery guidance. No input validation, no enum constraints, no parameter type constraints beyond basic JSON Schema. The search_web tool is the strongest (detailed description, multiple providers documented), but others are skeletal. Most tools would score 30-45 individually.
Export workflow to n8n format.
Generate agents.yaml configuration for a topic.
Run a PraisonAI agent with a prompt.
Auto-generate and run agents for a topic.
Run task with agent handoff/delegation.
Run deep research on a topic using DeepResearchAgent.
Search the web using multiple providers with automatic fallback. Tries each search provider in order until one succeeds. Provider priority: 1. Tavily (requires TAVILY_API_KEY) 2. Exa (requires EXA_API_KEY) 3. You.com (requires YDC_API_KEY) 4. DuckDuckGo (no API key needed) 5. SearxNG (requires SEARXNG_URL)
Tool names use generic verbs ('run_agent', 'run_research', 'run_auto_agents') that do not clearly distinguish intent. LLMs cannot infer from the name alone whether to call run_agent or run_research for a given task.
Output schemas are not documented. Callers cannot predict response structure (e.g., does run_agent return 'result' or 'output'? Is it a string or object?). This forces LLMs to guess and wastes tokens on trial-and-error.
Parameter descriptions are minimal (10-30 chars, vs. 72-char baseline). E.g., 'The task or question for the agent' is adequate, but 'Research topic' and 'Topic to generate agents for' lack context on format, length, or examples.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
Create a workflow definition.
Create workflow from YAML configuration.
Run a multi-step workflow.
Error handling returns generic dicts with 'error' key but no recovery guidance. E.g., 'praisonaiagents not installed' tells the LLM nothing about what to do next. No categorization of retryable vs. fatal errors.
No input validation or enum constraints. E.g., 'model' parameter in run_agent accepts any string; no validation that it is a valid LLM model. 'providers' in search_web lists options in description but not as an enum constraint.