Praison AI
Exposes PraisonAI's multi-agent framework as an MCP server, so any MCP client can run agents, kick off research, execute code, and search the web through 64 built-in tools.
About
praisonai-mcp is the MCP-facing counterpart to the PraisonAI agent framework: instead of writing Python to define agents, an MCP client like Claude Desktop or Cursor gets direct tool access to run them. That means running a single agent against a prompt, kicking off a multi-step research task, auto-generating agents for a job you describe, or executing a whole PraisonAI workflow, all as one MCP tool call rather than a script you have to run separately.
Underneath, it inherits PraisonAI's broader feature set: memory (short and long-term, file- or graph-based), RAG over PDFs and other documents, and support for 100+ LLMs across 24+ providers with automatic cost-based model routing. The MCP server's other 60-plus tools round this out with real utilities, including multiple web-search providers with fallback, page scraping, Python and shell execution, git commits, and repository analysis, so an agent session isn't limited to conversation, it can actually go do things.
Key features
- 64 built-in tools covering PraisonAI's agent, research, search, and code-execution capabilities
- run_agent / run_research / run_auto_agents / workflow_run tools for one-call agent execution
- Search tools with automatic provider fallback (Tavily, Exa, DuckDuckGo, Wikipedia, arXiv)
- Code and shell execution plus git and repository-analysis tools for development tasks
- Inherits PraisonAI's memory, RAG, and 100+-LLM/24+-provider support
- Installable via uvx or pip, and registered in the MCP Registry
Use cases
- Running a research task from inside an MCP chat client instead of scripting PraisonAI directly
- Auto-generating a set of task-specific agents for a one-off job described in natural language
- Having an agent execute Python or shell commands and commit the result via git, all through MCP tools
- Scraping and summarizing web pages as part of a larger agent workflow triggered from Claude or Cursor
Available tools
run_agent
Executes a single PraisonAI agent against a given prompt.
run_research
Runs a multi-step research task on a topic.
run_auto_agents
Auto-generates agents suited to a described task.
workflow_run
Executes a predefined multi-step PraisonAI workflow.
search_web
Unified web search with automatic fallback across providers.
run_python
Executes Python code and returns the result.
Frequently asked questions
Is this the same thing as the main PraisonAI Python framework?
It's built on it. praisonai-mcp exposes PraisonAI's agents and workflows as MCP tools, so an MCP client can use them without writing PraisonAI Python code directly.
Does it require its own API keys, or does it reuse PraisonAI's provider config?
It uses the same provider setup as PraisonAI generally. You configure whichever of the 100+ supported LLM providers you want to use, plus keys for any search providers you enable.