grip-sdk

The official Python client and development library for Grip AI — Dynamic multi-provider search, deep reasoning token streaming, and conversational sessions.

Overview

grip-sdk is the official, high-performance developer library for Grip AI. It allows Python applications and AI agents to query live grounded web information, stream reasoning tokens in real-time, maintain multi-turn conversational history, and invoke custom tools or Model Context Protocol (MCP) servers.

Zero external microservices or Docker dependencies required — grip-sdk is a pure, self-contained Python package tested across Python 3.10, 3.11, 3.12, and 3.13.

Available directly on PyPI as grip-sdk and maintained on GitHub at theshoaibme/grip-sdk.

Installation

pip install grip-sdk

Quick start

Instantiate the synchronous Grip client and run an AI search query with citation grounding:

from grip import Grip

client = Grip()
response = client.search.query("What are the latest breakthroughs in fusion energy?")

print("Answer:", response.answer)
print("\nSources:")
for src in response.sources:
    print(f"- {src.title}: {src.link}")

Token streaming & reasoning

Stream tokens directly as they are generated, including deep reasoning and thinking events:

from grip import Grip

client = Grip()

for chunk in client.search.stream("Explain quantum entanglement for beginners"):
    if chunk.event == "thinking":
        print(f"[Thinking]: {chunk.thinking}")
    elif chunk.event == "delta":
        print(chunk.delta, end="", flush=True)
    elif chunk.event == "done":
        print("\n[Complete]")

Multi-turn conversational sessions

Maintain persistent conversational memory and thread context across consecutive turns:

from grip import Grip

client = Grip()

# Create or resume a session
session = client.sessions.create(title="Physics Assistant")

# First turn
res1 = session.ask("What is dark matter?")
print(res1.answer)

# Follow-up turn retains context
res2 = session.ask("How do scientists try to detect it?")
print(res2.answer)

# List historical turns
history = session.get_history()
print(f"Total turns in session: {len(history)}")

Asynchronous client (AsyncGrip)

For asynchronous workflows in FastAPI, asyncio, or agent loops, use AsyncGrip:

import asyncio
from grip import AsyncGrip

async def main():
    async with AsyncGrip() as client:
        # Async query
        res = await client.search.query("What is asyncio in Python?")
        print(res.answer)

        # Async streaming
        async for chunk in client.search.stream("Explain event loops"):
            if chunk.event == "delta":
                print(chunk.delta, end="", flush=True)

asyncio.run(main())

Dynamic multi-provider search

grip-sdk provides automatic routing and dynamic dispatch across multiple providers without code changes:

  • google / google_ai — Google AI Mode stealth browser automation
  • openai — OpenAI AI search integration
  • chatgpt_web — ChatGPT web interface provider with reasoning models
  • gemini_web — Gemini web interface provider
  • grip / grip_ai — Grip AI native intelligent auto-router
# Target specific provider or let Grip auto-route
client = Grip(provider="google_ai")
res = client.search.query("Latest NASA missions")

Agentic tools & MCP integration

Build agentic workflows with custom tool registration, behavioral rules, and external Model Context Protocol (MCP) servers:

from grip import Grip
from grip.agent import Agent

client = Grip()
agent = Agent(client=client)

# Connect an external MCP server via stdio
agent.connect_mcp(command="npx", args=["-y", "@modelcontextprotocol/server-filesystem", "/tmp"])

# Execute agent run with reasoning and tools
result = agent.run("Analyze the latest dataset and summarize key takeaways")
print(result.output)

PyPI Package

View release history, wheels, and download statistics on PyPI.

GitHub Repository

Source code, issues, examples, and tests on the official repository.

CLI Package

Explore the standalone Grip command-line interface.

MCP Configuration

Configure Grip MCP server in Cursor, Claude Desktop, and VS Code.