11![ MCP Toolbox Logo] ( https://raw.githubusercontent.com/googleapis/genai-toolbox/main/logo.png )
2- # MCP Toolbox LangChain SDK
2+ # MCP Toolbox LlamaIndex SDK
33
44This SDK allows you to seamlessly integrate the functionalities of
5- [ Toolbox] ( https://github.com/googleapis/genai-toolbox ) into your LangChain LLM
5+ [ Toolbox] ( https://github.com/googleapis/genai-toolbox ) into your LlamaIndex LLM
66applications, enabling advanced orchestration and interaction with GenAI models.
77
88<!-- TOC ignore:true -->
@@ -15,10 +15,7 @@ applications, enabling advanced orchestration and interaction with GenAI models.
1515- [ Loading Tools] ( #loading-tools )
1616 - [ Load a toolset] ( #load-a-toolset )
1717 - [ Load a single tool] ( #load-a-single-tool )
18- - [ Use with LangChain] ( #use-with-langchain )
19- - [ Use with LangGraph] ( #use-with-langgraph )
20- - [ Represent Tools as Nodes] ( #represent-tools-as-nodes )
21- - [ Connect Tools with LLM] ( #connect-tools-with-llm )
18+ - [ Use with LlamaIndex] ( #use-with-llamaindex )
2219- [ Manual usage] ( #manual-usage )
2320- [ Authenticating Tools] ( #authenticating-tools )
2421 - [ Supported Authentication Mechanisms] ( #supported-authentication-mechanisms )
@@ -38,41 +35,48 @@ applications, enabling advanced orchestration and interaction with GenAI models.
3835## Installation
3936
4037``` bash
41- pip install toolbox-langchain
38+ pip install toolbox-llamaindex
4239```
4340
4441## Quickstart
4542
4643Here's a minimal example to get you started using
47- [ LangGraph] ( https://langchain-ai.github.io/langgraph/reference/prebuilt/#langgraph.prebuilt.chat_agent_executor.create_react_agent ) :
44+ # TODO: add link
45+ [ LlamaIndex] ( ) :
4846
4947``` py
50- from toolbox_langchain import ToolboxClient
51- from langchain_google_vertexai import ChatVertexAI
52- from langgraph.prebuilt import create_react_agent
48+ import asyncio
5349
54- toolbox = ToolboxClient(" http://127.0.0.1:5000"
55- tools = toolbox.load_toolset()
50+ from llama_index.llms.google_genai import GoogleGenAI
51+ from llama_index.core.agent.workflow import AgentWorkflow
52+
53+ from toolbox_llamaindex import ToolboxClient
5654
57- model = ChatVertexAI(model = " gemini-1.5-pro-002"
58- agent = create_react_agent(model, tools)
55+ async def run_agent ():
56+ toolbox = ToolboxClient(" http://127.0.0.1:5000"
57+ tools = toolbox.load_toolset()
5958
60- prompt = " How's the weather today?"
59+ vertex_model = GoogleGenAI(
60+ model = " gemini-1.5-pro"
61+ vertexai_config = {" project" " project-id" " location" " us-central1"
62+ )
63+ agent = AgentWorkflow.from_tools_or_functions(
64+ tools,
65+ llm = vertex_model,
66+ system_prompt = " You are a helpful assistant."
67+ )
68+ response = await agent.run(user_msg = " Get some response from the agent."
69+ print (response)
6170
62- for s in agent.stream({" messages" " user" stream_mode = " values"
63- message = s[" messages" - 1 ]
64- if isinstance (message, tuple ):
65- print (message)
66- else :
67- message.pretty_print()
71+ asyncio.run(run_agent())
6872```
6973
7074## Usage
7175
7276Import and initialize the toolbox client.
7377
7478``` py
75- from toolbox_langchain import ToolboxClient
79+ from toolbox_llamaindex import ToolboxClient
7680
7781# Replace with your Toolbox service's URL
7882toolbox = ToolboxClient(" http://127.0.0.1:5000"
@@ -102,85 +106,63 @@ tool = toolbox.load_tool("my-tool")
102106Loading individual tools gives you finer-grained control over which tools are
103107available to your LLM agent.
104108
105- ## Use with LangChain
109+ ## Use with LlamaIndex
106110
107111LangChain's agents can dynamically choose and execute tools based on the user
108112input. Include tools loaded from the Toolbox SDK in the agent's toolkit:
109113
110114``` py
111- from langchain_google_vertexai import ChatVertexAI
115+ from llama_index.llms.google_genai import GoogleGenAI
116+ from llama_index.core.agent.workflow import AgentWorkflow
112117
113- model = ChatVertexAI(model = " gemini-1.5-pro-002"
118+ vertex_model = GoogleGenAI(
119+ model = " gemini-1.5-pro"
120+ vertexai_config = {" project" " project-id" " location" " us-central1"
121+ )
114122
115123# Initialize agent with tools
116- agent = model.bind_tools(tools)
117-
118- # Run the agent
119- result = agent.invoke(" Do something with the tools"
120- ```
121-
122- ## Use with LangGraph
123-
124- Integrate the Toolbox SDK with LangGraph to use Toolbox service tools within a
125- graph-based workflow. Follow the [ official
126- guide] ( https://langchain-ai.github.io/langgraph/ ) with minimal changes.
127-
128- ### Represent Tools as Nodes
129-
130- Represent each tool as a LangGraph node, encapsulating the tool's execution within the node's functionality:
131-
132- ``` py
133- from toolbox_langchain import ToolboxClient
134- from langgraph.graph import StateGraph, MessagesState
135- from langgraph.prebuilt import ToolNode
136-
137- # Define the function that calls the model
138- def call_model (state : MessagesState):
139- messages = state[' messages'
140- response = model.invoke(messages)
141- return {" messages" # Return a list to add to existing messages
142-
143- model = ChatVertexAI(model = " gemini-1.5-pro-002"
144- builder = StateGraph(MessagesState)
145- tool_node = ToolNode(tools)
146-
147- builder.add_node(" agent"
148- builder.add_node(" tools"
124+ agent = AgentWorkflow.from_tools_or_functions(
125+ tools,
126+ llm = vertex_model,
127+ system_prompt = " You are a helpful assistant."
128+ )
129+
130+ # Query the agent
131+ response = await agent.run(user_msg = " Get some response from the agent."
132+ print (response)
149133```
150134
151- ### Connect Tools with LLM
135+ ### Maintain state
152136
153- Connect tool nodes with LLM nodes. The LLM decides which tool to use based on
154- input or context. Tool output can be fed back into the LLM:
137+ To maintain state for the agent, add context as follows:
155138
156139``` py
157- from typing import Literal
158- from langgraph.graph import END , START
159- from langchain_core.messages import HumanMessage
160-
161- # Define the function that determines whether to continue or not
162- def should_continue (state : MessagesState) -> Literal[" tools" END ]:
163- messages = state[' messages'
164- last_message = messages[- 1 ]
165- if last_message.tool_calls:
166- return " tools" # Route to "tools" node if LLM makes a tool call
167- return END # Otherwise, stop
168-
169- builder.add_edge(START , " agent"
170- builder.add_conditional_edges(" agent"
171- builder.add_edge(" tools" ' agent'
172-
173- graph = builder.compile()
174-
175- graph.invoke({" messages" content = " Do something with the tools"
140+ from llama_index.core.agent.workflow import AgentWorkflow
141+ from llama_index.core.workflow import Context
142+ from llama_index.llms.google_genai import GoogleGenAI
143+
144+ vertex_model = GoogleGenAI(
145+ model = " gemini-1.5-pro"
146+ vertexai_config = {" project" " twisha-dev" " location" " us-central1"
147+ )
148+ agent = AgentWorkflow.from_tools_or_functions(
149+ tools,
150+ llm = vertex_model,
151+ system_prompt = " You are a helpful assistant"
152+ )
153+
154+ # Save memory in agent context
155+ ctx = Context(agent)
156+ response = await agent.run(user_msg = " Give me some response." ctx = ctx)
157+ print (response)
176158```
177159
178160## Manual usage
179161
180- Execute a tool manually using the ` invoke ` method:
162+ Execute a tool manually using the ` call ` method:
181163
182164``` py
183- result = tools[0 ].invoke ({" name" " Alice" " age" 30 })
165+ result = tools[0 ].call ({" name" " Alice" " age" 30 })
184166```
185167
186168This is useful for testing tools or when you need precise control over tool
@@ -250,7 +232,7 @@ auth_tools = toolbox.load_toolset(auth_tokens={"my_auth": get_auth_token})
250232
251233``` py
252234import asyncio
253- from toolbox_langchain import ToolboxClient
235+ from toolbox_llamaindex import ToolboxClient
254236
255237async def get_auth_token ():
256238 # ... Logic to retrieve ID token (e.g., from local storage, OAuth flow)
@@ -261,7 +243,7 @@ toolbox = ToolboxClient("http://127.0.0.1:5000")
261243tool = toolbox.load_tool(" my-tool"
262244
263245auth_tool = tool.add_auth_token(" my_auth"
264- result = auth_tool.invoke ({" input" " some input"
246+ result = auth_tool.call ({" input" " some input"
265247print (result)
266248```
267249
@@ -329,7 +311,7 @@ use the asynchronous interfaces of the `ToolboxClient`.
329311
330312``` py
331313import asyncio
332- from toolbox_langchain import ToolboxClient
314+ from toolbox_llamaindex import ToolboxClient
333315
334316async def main ():
335317 toolbox = ToolboxClient(" http://127.0.0.1:5000"
0 commit comments