chatbot / src /Agentic_System /Google_Docs_Agent.py
jawadsaghir12's picture
new update
a66d4bd
"""
Google Docs Agent – specialized agent with only Docs MCP tools.
Uses MCPServerStdio (local subprocess) with create_static_tool_filter so the
agent sees *only* document-related tools. Zero network overhead.
"""
import asyncio
import logging
from openai import AsyncOpenAI
from agents import Agent, Runner, OpenAIChatCompletionsModel
from agents.model_settings import ModelSettings
try:
from .google_mcp_config import (
LONGCAT_API_KEY, LONGCAT_BASE_URL, MODEL_NAME,
DOCS_TOOLS, create_google_mcp_server, USER_GOOGLE_EMAIL,
)
except ImportError:
from google_mcp_config import (
LONGCAT_API_KEY, LONGCAT_BASE_URL, MODEL_NAME,
DOCS_TOOLS, create_google_mcp_server, USER_GOOGLE_EMAIL,
)
logger = logging.getLogger(__name__)
SYSTEM_PROMPT = """\
You are a specialized Google Docs assistant. You can create, read, modify,
and manage Google Documents using the available tools.
Capabilities:
- **Search & list** documents by title or folder
- **Read** full document content or inspect its structural elements
- **Create** new documents with initial content
- **Edit** text β€” insert, modify, find-and-replace, batch updates
- **Insert elements** β€” images, tables (with data), page breaks
- **Headers & footers** β€” create and update
- **Paragraph styling** β€” alignment, spacing, indentation, named styles
- **Export** documents to PDF
- **Comments** β€” read, create, reply to, and resolve
Rules:
1. The user's Google email is provided in the query β€” use it for every
tool call in the `user_google_email` parameter. NEVER ask the user
for their email; it is always supplied.
2. When inserting content, be mindful of the document index positions.
Use `inspect_doc_structure` first when modifying existing docs to
understand the current element layout.
3. For batch updates, send all changes in a single `batch_update_doc`
call when possible for atomic application.
4. After modifications, briefly confirm what was changed.
"""
class GoogleDocsAgent:
"""Thin wrapper around the OpenAI Agent SDK wired to Google Docs tools."""
def __init__(self, model: str = MODEL_NAME):
self.model = model
self._client = AsyncOpenAI(
api_key=LONGCAT_API_KEY,
base_url=LONGCAT_BASE_URL,
timeout=30.0,
)
# ── factory helpers ──────────────────────────────────────────────────
def _create_mcp_server(self):
"""Spawn a local MCP subprocess with only Docs tools loaded."""
return create_google_mcp_server(service="docs", tool_names=DOCS_TOOLS)
def _create_agent(self, mcp_server) -> Agent:
return Agent(
name="Google Docs Agent",
instructions=SYSTEM_PROMPT,
mcp_servers=[mcp_server],
model=OpenAIChatCompletionsModel(
model=self.model,
openai_client=self._client,
),
model_settings=ModelSettings(tool_choice="auto"),
)
# ── public API ───────────────────────────────────────────────────────
async def run(self, query: str) -> str:
"""Spawn MCP connection, run a single query, then clean up."""
mcp_server = self._create_mcp_server()
async with mcp_server:
agent = self._create_agent(mcp_server)
logger.info("Google Docs MCP connected – agent ready")
result = await Runner.run(agent, input=query)
return result.final_output
# ─── CLI entry point ──────────────────────────────────────────────────────────
async def main():
agent = GoogleDocsAgent()
resp = await agent.run(
"Search for documents with 'meeting notes' in the title. "
"My email is user@example.com"
)
print("Agent Response:\n", resp)
if __name__ == "__main__":
from dotenv import load_dotenv, find_dotenv
load_dotenv(find_dotenv())
asyncio.run(main())