# PraisonAI [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/MervinPraison/PraisonAI  
**GitHub Stars:** 9330  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/praisonai

## Description
AI Agents Framework with Self Reflection and MCP support

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "praisonai": {
    "command": "npx",
    "args": ["-y","@modelcontextprotocol/server-filesystem"],
    "env": {
      "OPENAI_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `OPENAI_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What PraisonAI does

PraisonAI provides Python APIs for creating agents that can research, plan, execute tasks, and produce structured output. Its scope ranges from individual agents to coordinated groups using agent flows, handoffs, routing, parallel execution, loops, and repeated tasks.

The framework also includes controls for the agent execution cycle. Developers can configure memory, knowledge sources, context compression, guardrails, human approval, hooks, sandbox execution, reflection, autonomy, and checks for stalled or repetitive behavior.

MCP support lets an agent use tools exposed by MCP servers. The README demonstrates combining a local Python tool with an MCP filesystem server, so PraisonAI can place MCP-based capabilities alongside application-defined functions.

## How it works

Agents are created in Python with the `Agent` class from `praisonaiagents`. An agent can receive instructions, a role, a goal, output preferences, memory, knowledge, context settings, and a tool list. Calling `agent.start(...)` begins execution for a user-provided task.

The framework describes execution as layered configuration:

- Prompt settings define the role, instructions, examples, and output format.
- Context settings manage memory, knowledge selection, compression, and handoff context.
- Harness settings provide tools, MCP connections, guardrails, approvals, hooks, and sandbox options.
- Loop settings control iterations, budgets, reflection, autonomy, and progress checks.
- Graph settings coordinate multiple agents through routing, parallel work, loops, and repetition.

A handoff can pass work to a specialist while limiting the inherited conversation and tools. Managed execution can run tools in Docker or run an entire agent through an Anthropic-backed option, as shown by the framework's configuration examples.

## Setup and configuration

Install the main package with pip:

```bash
pip install praisonai
```

For the lightweight agent SDK, install `praisonaiagents` instead or in addition, then configure an OpenAI API key:

```bash
pip install praisonaiagents
export OPENAI_API_KEY="your-api-key"
```

The project also provides an installation script:

```bash
curl -fsSL https://praison.ai/install.sh | bash
```

A basic agent can be defined with Python:

```python
from praisonaiagents import Agent

agent = Agent(instructions="You are a senior data analyst.")
agent.start("Analyze the top 3 tech trends of 2026")
```

## Tools and capabilities

- Connect agents to MCP servers with `MCP(...)`.
- Register Python functions as agent tools with the `@tool` decorator.
- Add memory, knowledge directories, and context summarization.
- Require approval before selected tools execute.
- Organize multiple agents with flows, routing, parallel work, loops, and handoffs.
- Configure execution limits, reflection, autonomy, and progress checks.

## Limitations and notes

PraisonAI is presented as a Python framework rather than a client-specific MCP application. The documented MCP example shows an agent consuming an MCP filesystem server; MCP server connection details depend on the command or configuration passed to `MCP(...)`. The example uses an OpenAI API key, and the framework also includes an option for running an agent through Anthropic.

_Full upstream README: https://allmcps.com/mcp/praisonai/readme_