PraisonAI MCP

Name: PraisonAI MCP
Author: Mervin Praison

Exposes the PraisonAI agent framework's agents, memory, knowledge and model tools over MCP

Official · Mervin Praison Python Local · stdio AI and ML Knowledge and memory

GitHub PyPI Website

79 popularity score ★ 9.3k ↓ 3.1k/mo

Connect your client

Local (PyPI)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio

Opens the app, which asks before it adds the server. Then replace each <NAME> in its settings with your value.

Or add it by command or config

It needs OPENAI_API_KEY. Replace each <NAME> with your value.

FirstTerminal

pip install "praisonai-mcp[all]"

Claude CodeTerminal

claude mcp add praisonai -e OPENAI_API_KEY=<OPENAI_API_KEY> -- praisonai-mcp serve --transport stdio

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "praisonai": {
      "command": "praisonai-mcp",
      "args": [
        "serve",
        "--transport",
        "stdio"
      ],
      "env": {
        "OPENAI_API_KEY": "<OPENAI_API_KEY>"
      }
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "praisonai": {
      "command": "praisonai-mcp",
      "args": [
        "serve",
        "--transport",
        "stdio"
      ],
      "env": {
        "OPENAI_API_KEY": "<OPENAI_API_KEY>"
      }
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "praisonai": {
      "type": "stdio",
      "command": "praisonai-mcp",
      "args": [
        "serve",
        "--transport",
        "stdio"
      ],
      "env": {
        "OPENAI_API_KEY": "<OPENAI_API_KEY>"
      }
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.praisonai]
command = "praisonai-mcp"
args = ["serve", "--transport", "stdio"]

[mcp_servers.praisonai.env]
OPENAI_API_KEY = "<OPENAI_API_KEY>"

Gemini CLITerminal

gemini mcp add --scope user -e OPENAI_API_KEY=<OPENAI_API_KEY> praisonai praisonai-mcp -- serve --transport stdio

Settings 3

`OPENAI_API_KEY` secret	Required: Key for the default OpenAI models used by the model-backed tools. `doctor` warns "No API keys configured" without any provider key.
`ANTHROPIC_API_KEY` secret	Optional: Optional Anthropic key, for Claude models named in a tool's model argument.
`GOOGLE_API_KEY` secret	Optional: Optional Google key, for Gemini models named in a tool's model argument.

What it does

PraisonAI MCP is the server package of the PraisonAI agent framework. You start it with `praisonai-mcp serve`, and it hands the framework's features to a client such as Claude Desktop or Cursor as MCP tools. From there the client can pass a task to a PraisonAI agent, run a workflow or a research job, make images, video or speech, transcribe audio, and keep memory and a knowledge base. It also manages rule files, todos and scheduled tasks, and can work with a provider's files, batches, vector stores and fine-tuning jobs. Alongside the tools come seven read-only resources and seven prompt templates.

Transport is stdio unless you pass `--transport http-stream`. In that mode it binds to 127.0.0.1 on port 8080 at `/mcp`, and `--host`, `--port` and `--endpoint` change that. With the full `praisonai` package installed the same commands also run as `praisonai mcp ...`.

The code registers 85 tools, all named `praisonai.<area>.<action>`, where the docs count 86. Most of them need more than this package. The model tools (chat completion, images, audio, embeddings, moderation, rerank, RAG, web search, guardrails), workflow runs and everything in the extended set (video, provider files, batches, vector stores, assistants, fine-tuning, OCR, containers, deploy) import their code from the separate `praisonai` package. Without it they come back with an error asking you to run `pip install praisonai` (capabilities.py, extended_capabilities.py, _wrapper_bridge.py). The agent, research, memory, knowledge, rules, todo and schedule tools run on this package alone.

The project README now talks up running offline on a local model. For this server, models are fixed per tool in code, and the defaults are OpenAI ones (gpt-4o-mini, dall-e-3, whisper-1 and so on). `agent.chat`, `agent.run` and `chat.completion` accept a `model` argument, so a client can name something like `ollama/llama3.2` there. `research.run` always calls gpt-4o, and no setting changes the default model server-wide.

What it needs

Runtime	Python 3.10+
Runs on	macOS, Windows, Linux
OpenAI	Paid planDefault provider for the model-backed tools. Anthropic or Google keys are optional, and agent and chat tools can take a local model name instead
Also	The praisonai package (pip install praisonai) for the model, workflow and extended tools

Install with `pip install "praisonai-mcp[all]"`, or `pip install "praisonai[mcp]"` for the whole framework. Add `pip install praisonai` if you installed the standalone package and want the model, workflow and extended tools. Set OPENAI_API_KEY for the default models. To use a local model with the agent and chat tools, have Ollama running and pass its model name in the tool's `model` argument. Provider calls are billed by the provider.

Limits

Limit Model, workflow, deploy and extended tools return an error unless the separate praisonai package is installed too
Not covered No server-wide model setting. Only agent.chat, agent.run and chat.completion take a model name, and research.run always uses gpt-4o
Limit Before PraisonAI PR #4311, knowledge, session, workflow, audio, OCR, a2a, containers, skills and realtime tools answered "not available". Upgrade older installs
Limit knowledge.clear returns an error naming the backend when that backend (for example Chroma) cannot reset
Limit The docs list an ocr.extract tool, but the registered name is praisonai.ocr
Cost Image, video, speech, batch and fine-tuning tools run on the provider account whose key is set and are billed there

Without the `praisonai` package co-installed, the extended tools (video, files, batches, vector stores, assistants, fine-tuning, OCR, containers, workflow.auto, deploy) return an error instead of working. Its docs note that knowledge, session, hooks, eval, workflow, audio, guardrails, OCR, a2a, containers, skills and realtime tools returned "not available" errors in releases before PR #4311; older installs should be upgraded. `knowledge.clear` reports an error when the configured backend (for example Chroma) cannot reset. The docs' tool list uses `ocr.extract`, but the registered tool is `praisonai.ocr`.

Who it suits

Suits developers already building with PraisonAI who want to drive its agents, workflows and memory from Claude Desktop, Cursor or another MCP client. In practice that means an OpenAI key and the full `praisonai` package, since most tools default to OpenAI models and import from it. Running fully on a local model only goes as far as the few tools that take a `model` argument. It puts 85 tools in front of the client at once, so someone who wants a single capability (only web search, only memory) will find a narrower server easier to reason about. Those who only need to serve a few custom Python tools can use PraisonAI's lighter `praisonai serve mcp` server, as the package page describes.

Tools 60 of 85

31 only read, 20 change things, 9 can delete or overwrite. A ? marks an optional parameter.

praisonai.agent.chat: Send a message to a general PraisonAI agent and get its reply; model is gpt-4o-mini unless givenmessage: stringmodel?: string
praisonai.agent.run: Have a general agent carry out a task, with gpt-4o-mini unless another model is giventask: stringmodel?: string
praisonai.research.run: Run a research agent on a topic, always on gpt-4otopic: stringdepth?: string
praisonai.workflow.run changes: Run a workflow passed in as YAML textyaml_content: string
praisonai.workflow.run_file changes: Run a workflow from a YAML file on diskfile_path: string
praisonai.workflow.auto changes: Generate a workflow for a topic in a chosen patterntopic: stringpattern?: string
praisonai.workflow.list: List workflow files found
praisonai.workflow.validate: Check a workflow YAML filefile_path: string
praisonai.workflow.show: Show a workflow file's configurationfile_path: string
praisonai.chat.completion: Get a single chat completion, with an optional system promptmessage: stringmodel?: stringsystem_prompt?: string
praisonai.images.generate changes: Create images from a prompt, dall-e-3 at 1024x1024 by defaultprompt: stringmodel?: stringsize?: stringn?: integer
praisonai.videos.generate changes: Create a short video from a prompt, openai/sora-2 for 4 seconds by defaultprompt: stringmodel?: stringduration?: integer
praisonai.audio.transcribe: Turn an audio file into text with whisper-1 by defaultfile_path: stringmodel?: stringlanguage?: string
praisonai.audio.speech changes: Turn text into spoken audio, tts-1 with the alloy voice by defaulttext: stringmodel?: stringvoice?: string
praisonai.embed.create: Compute embeddings for texttext: stringmodel?: string
praisonai.moderate.check: Run content moderation on texttext: string
praisonai.rerank: Reorder documents by relevance to a query, top 5 by defaultquery: stringdocuments: arraymodel?: stringtop_n?: integer
praisonai.search: Search the web for a queryquery: string
praisonai.ocr: Pull text out of an image with a vision model, gpt-4o by defaultimage_path: stringmodel?: string
praisonai.guardrails.check: Check text against a named guardrailtext: stringguardrail_name?: string
praisonai.memory.show: Show stored memory, optionally for one useruser_id?: string
praisonai.memory.add changes: Save an item to memory with optional metadatacontent: stringmetadata?: string
praisonai.memory.search: Search stored memory, 10 results by defaultquery: stringlimit?: integeruser_id?: string
praisonai.memory.clear deletes: Wipe stored memory
praisonai.knowledge.add changes: Add a file path, URL or raw text to the knowledge basesource: string
praisonai.knowledge.query: Ask the knowledge base a question, 5 results by defaultquery: stringlimit?: integer
praisonai.knowledge.list: List knowledge sources
praisonai.knowledge.clear deletes: Empty the knowledge base, or say which backend refused
praisonai.knowledge.stats: Report knowledge base statistics
praisonai.rag.query: Answer a question by retrieval over a stored collectionquery: stringcollection?: string
praisonai.rules.list: List rule files in ~/.praison/rules
praisonai.rules.show: Read one rule filerule_name: string
praisonai.rules.create deletes: Write a rule file, replacing one of the same namerule_name: stringcontent: string
praisonai.rules.delete deletes: Remove a rule filerule_name: string
praisonai.todo.list: List todo items
praisonai.todo.add changes: Add a todo item with a priority, medium by defaultcontent: stringpriority?: string
praisonai.todo.complete changes: Mark a todo donetodo_id: string
praisonai.todo.delete deletes: Remove a todo itemtodo_id: string
praisonai.schedule.list: List scheduled tasks
praisonai.schedule.add changes: Schedule a workflow file on a cron expressiontask_name: stringcron: stringworkflow_path: string
praisonai.schedule.remove deletes: Cancel a scheduled tasktask_name: string
praisonai.session.list: List agent sessions
praisonai.session.info: Show details of one sessionsession_id: string
praisonai.session.delete deletes: Delete a sessionsession_id: string
praisonai.files.create changes: Upload a local file to the model providerfile_path: stringpurpose?: string
praisonai.files.list: List files uploaded to the providerpurpose?: string
praisonai.files.delete deletes: Delete an uploaded file at the providerfile_id: string
praisonai.batches.create changes: Start a batch job from an uploaded fileinput_file_id: stringendpoint?: stringcompletion_window?: string
praisonai.batches.cancel changes: Cancel a running batch jobbatch_id: string
praisonai.vector_stores.create changes: Create a provider vector store from uploaded filesname: stringfile_ids?: array
praisonai.vector_stores.search: Search a vector store, 10 results by defaultvector_store_id: stringquery: stringmax_results?: integer
praisonai.assistants.create changes: Create an assistant at the providername: stringinstructions: stringmodel?: stringtools?: array
praisonai.fine_tuning.create changes: Start a fine-tuning job on a training filetraining_file: stringmodel?: stringsuffix?: string
praisonai.containers.create changes: Create a code container, python:3.11 by defaultname: stringimage?: string
praisonai.containers.file_write deletes: Write a file inside a containercontainer_id: stringfile_path: stringcontent: string
praisonai.a2a.send changes: Send a message to another agent's A2A addressagent_url: stringmessage: stringtask_id?: string
praisonai.skills.load: Load a skill from a pathskill_path: string
praisonai.tools.search: Search the framework's available toolsquery: string
praisonai.mcp_config.enable changes: Turn on a saved MCP server configurationconfig_name: string
praisonai.doctor.env: Report which provider keys and settings are present

Resources 7

praisonai://memory/sessions: Sessions in the session store
praisonai://workflows: Workflow files in the current directory
praisonai://tools: Tools available to agents
praisonai://agents: Agent configurations
praisonai://knowledge/sources: Sources in the knowledge base
praisonai://config: Current configuration
praisonai://mcp/status: Status of this MCP server

Prompts 7

deep-research: Builds a prompt for researching a topic in depth
code-review: Builds a prompt for reviewing code
workflow-auto: Builds a prompt for generating a workflow
guardrail-check: Builds a prompt for checking content safety
context-engineering: Builds a prompt that shapes context for a model
eval-criteria: Builds criteria for judging agent output
agent-instructions: Builds instructions for an agent in a given role

Popularity

79 popularity score, on our list (the bar is 36)

★ 9.3k ↓ 3.1k/mo

GitHub stars	9,325	47.6 points
PyPI downloads, last month	3,096	23.1 points
Official server	Yes	8 points

Measured 2026-11-07. How the score works.

Where it is published

Code	mervinpraison/praisonai ★ 9.3k
PyPI	praisonai-mcp

What it can reach

Runs	On your machine
Reaches	Files on your machine, The internet, Your account at a service
Changes things	Yes: it can write, send, create or delete
Sign-in	An API key or token
Licence	MIT (GitHub)
Last code change	2026-11-05

Runs on your machine. Tools send prompts, files and audio to the model providers whose keys you set (OpenAI by default) and can upload or delete files, start batches and fine-tuning jobs on that account. Rule tools read and write files under ~/.praison/rules; since 4.6.34 rule names with separators, `..` or a leading dot are refused (GHSA-9mqq-jqxf-grvw). Memory, knowledge, session and todo tools change local stores, and knowledge.add can fetch URLs. Over HTTP it listens on 127.0.0.1:8080 by default with origin checks; `--api-key` adds a Bearer token, and binding elsewhere needs `--allowed-origins`.

Over time

GitHub stars 9.3k +87 (+1 %)

9.3k9.2k

14 Oct 20267 Nov 2026

PyPI downloads a month 3.1k +3.0k (+2793 %)

4.4k107

2 Jun 20267 Nov 2026

Before the site measured it, read from pypistats' daily download counts.

Popularity score 79 +0

7979

14 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-11-06.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![PraisonAI MCP on EveryMCPServer](https://everymcpserver.com/badge/praisonai-mcp.svg)](https://everymcpserver.com/servers/praisonai-mcp)

Connect your client

Add in one click

Or add it by command or config

Settings 3

What it does

What it needs

Limits

Who it suits

Tools 60 of 85

Resources 7

Prompts 7

Similar servers

Ruflo

Clawdi

llmwiki

Bike4Mind

Engram

Memanto MCP Server