Documentation Index
Fetch the complete documentation index at: https://www.klavis.ai/docs/llms.txt
Use this file to discover all available pages before exploring further.
This is the legacy approach to MCP servers. We recommend using Strata MCP Server for better tool management and context window optimization.
- UI
- API
- Open Source
Open Dashboard
Go to your Dashboard.
Create a server instance
curl -X POST "https://api.klavis.ai/mcp-server/instance/create" \
-H "Authorization: Bearer YOUR_KLAVIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"serverName": "Gmail",
"userId": "user123"
}'
Response Information: The API returns:
serverUrl: The URL you’ll use to connect your MCP client to the individual MCP ServeroauthUrl: Authorization link if the service requires OAuth authentication
API Reference
Full Individual MCP Server endpoints
Authenticate
Copy and paste the OAuth URL into your web browser
Authentication Methods:
- API Key: See API Key authentication guide for details.
- OAuth: See OAuth authentication guide for details.
🎉 Your MCP Server URL is ready to use! Once authentication is complete, you can use your MCP server URL with any MCP-compatible client.
(optional) Connect to your AI application
- LangChain
- LlamaIndex
- CrewAI
- AutoGen
import os
import asyncio
import webbrowser
from klavis import Klavis
from klavis.types import McpServerName
from langchain_openai import ChatOpenAI
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from dotenv import load_dotenv
load_dotenv()
async def main() -> None:
klavis_client = Klavis(api_key=os.getenv("KLAVIS_API_KEY"))
# Step 1: Create a single MCP server (e.g., Gmail)
response = klavis_client.mcp_server.create_server_instance(
server_name=McpServerName.GMAIL,
user_id="demo_user",
)
# Step 2: Handle OAuth authorization if needed
if hasattr(response, 'oauth_url') and response.oauth_url:
webbrowser.open(response.oauth_url)
input("Press Enter after completing OAuth authorization...")
# Step 3: Create LangChain Agent with MCP Tools
mcp_client = MultiServerMCPClient({
"gmail": {
"transport": "streamable_http",
"url": response.server_url,
}
})
# Get all available tools from the server
tools = await mcp_client.get_tools()
# Setup LLM
llm = ChatOpenAI(model="gpt-4o-mini", api_key=os.getenv("OPENAI_API_KEY"))
# Step 4: Create LangChain agent with MCP tools
agent = create_react_agent(
model=llm,
tools=tools,
prompt=(
"You are a helpful assistant that can use MCP tools. "
),
)
# Step 5: Invoke the agent
result = await agent.ainvoke({
"messages": [{"role": "user", "content": "Search my inbox for unread emails and summarize."}],
})
# Print only the final AI response content
print(result["messages"][-1].content)
if __name__ == "__main__":
asyncio.run(main())
import os
import asyncio
import webbrowser
from klavis import Klavis
from klavis.types import McpServerName
from llama_index.llms.openai import OpenAI
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.tools.mcp import BasicMCPClient
from llama_index.tools.mcp import (
aget_tools_from_mcp_url,
)
from dotenv import load_dotenv
load_dotenv()
async def main() -> None:
klavis_client = Klavis(api_key=os.getenv("KLAVIS_API_KEY"))
# Step 1: Create a single MCP server (e.g., Gmail)
response = klavis_client.mcp_server.create_server_instance(
server_name=McpServerName.GMAIL,
user_id="demo_user",
)
# Step 2: Handle OAuth authorization if needed
if hasattr(response, 'oauth_url') and response.oauth_url:
webbrowser.open(response.oauth_url)
input("Press Enter after completing OAuth authorization...")
# Step 3: Create LlamaIndex Agent with MCP Tools
tools = await aget_tools_from_mcp_url(
response.server_url,
client=BasicMCPClient(response.server_url)
)
# Setup LLM
llm = OpenAI(model="gpt-4o-mini", api_key=os.getenv("OPENAI_API_KEY"))
# Step 4: Create LlamaIndex agent with MCP tools
agent = FunctionAgent(
name="gmail_agent",
description="Agent using Gmail MCP tools",
tools=tools,
llm=llm,
system_prompt=(
"You are a helpful assistant that can use MCP tools. "
),
)
# Step 5: Invoke the agent
result = await agent.run(
"Search my inbox for unread emails and summarize."
)
# Print the response
print(result)
if _name_ == "_main_":
asyncio.run(main())
Coming soon
import os
import asyncio
import webbrowser
from dotenv import load_dotenv
from klavis import Klavis
from klavis.types import McpServerName
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.ui import Console
from autogen_core import CancellationToken
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import StreamableHttpMcpToolAdapter, StreamableHttpServerParams
from autogen_ext.tools.mcp import mcp_server_tools
load_dotenv()
async def main() -> None:
klavis_client = Klavis(api_key=os.getenv("KLAVIS_API_KEY"))
# Create MCP server instance
response = klavis_client.mcp_server.create_server_instance(
server_name=McpServerName.GMAIL,
user_id="demo_user",
)
# Handle OAuth authorization if required
if getattr(response, "oauth_url", None):
webbrowser.open(response.oauth_url)
input("Press Enter after completing OAuth authorization...")
server_params = StreamableHttpServerParams(
url=response.server_url,
timeout=30.0,
sse_read_timeout=300.0,
terminate_on_close=True,
)
adapters = await mcp_server_tools(server_params)
model_client = OpenAIChatCompletionClient(model="gpt-4")
agent = AssistantAgent(
name="MailAI",
model_client=model_client,
tools=adapters,
system_message="You are a helpful Gmail assistant.",
)
await Console(
agent.run_stream(
task="Find My Latest Emails",
cancellation_token=CancellationToken()
)
)
if __name__ == "__main__":
asyncio.run(main())
Visit https://github.com/Klavis-AI/klavis/mcp_servers to view the source code and find more information
Run the server locally or in your infra
docker run -p 5000:5000 ghcr.io/klavis-ai/gmail-mcp-server:latest
Browse all available MCP server Docker images at GitHub Packages
Point your MCP client to the URL
Use the local URL (for example, http://localhost:5000) in your client or aggregator.
For the recommended approach with better tool management and progressive discovery, see Strata MCP Server.