Summary
An agent framework's whole toolbox, reachable from a client that is not it.
Research, crawling, memory, planning and agent delegation arrive as one set, and the unified search falls back across providers, so a research task keeps working when one key is missing or one provider is rate-limited.
What it is
An MCP front end for PraisonAI. Agent execution, multi-step workflows, a set of web search and crawl providers, and persistent memory and knowledge stores are all exposed as tools.
What you get
- Agents run from a prompt, generated automatically for a task, or handed off between each other, with the agents.yaml produced for you
- Multi-step workflows created, run, built from YAML, and exported to n8n
- Unified web search with automatic fallback across providers, plus direct access to Tavily, Exa, DuckDuckGo, SearXNG, You.com, Wikipedia and arXiv
- Pages scraped and sites crawled — Crawl4AI extraction and scraping, link extraction, and deep research as a single call
- A memory store that is added to, searched, listed and cleared, and that can extract memories from a conversation on its own
- A knowledge base kept separate from memory, added to and searched
- Repositories analysed, and plans created and executed as their own step
Requirements
Uvx praisonai-mcp, or pip install praisonai-mcp. The agent and workflow tools run PraisonAI locally; the search tools each need a key from their own provider — Tavily, Exa and the rest — and the provider list is readable, so you can see which are actually available.
Setup effort
One command plus a key — uvx praisonai-mcp, then supply credentials