Tool call fails in LangGraph Studio because messages stream mode cannot be disabled

[KathyDinh]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangGraph.js documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangGraph.js rather than my code.
• The bug is not resolved by updating to the latest stable version of LangGraph (or the specific integration package).

Example Code

TypeScript code for the graph router.ts is as follows:

// Tool
const multiplyTool = tool(
  async ({ a, b }: { a: number; b: number }) => {
    console.log(`Executing multiply tool: ${a}*${b}`);
    return a * b;
  },
  {
    name: "multiply",
    description: "Multiplies a and b.",
    schema: z.object({
      a: z.number().describe("first number"),
      b: z.number().describe("second number"),
    }),
  }
);

// LLM with bound tool
const llm = new ChatAnthropic({ model: "claude-3-5-haiku-latest", temperature: 0 });
const tools = [multiplyTool];
const llmWithTools = llm.bindTools(tools);

// Node
async function toolCallingLLM(state: typeof MessagesAnnotation.State) {
  const result = await llmWithTools.invoke(state.messages);

  return { messages: [result] };
}

enum NodeNames {
  TOOL_CALLING_LLM = "tool_calling_llm",
  TOOLS = "tools"
}

// Build graph
const builder = new StateGraph<typeof MessagesAnnotation.spec, any, any, NodeNames>(MessagesAnnotation);
builder.addNode(NodeNames.TOOL_CALLING_LLM, toolCallingLLM);
builder.addNode(NodeNames.TOOLS, new ToolNode(tools));
builder.addEdge(START, NodeNames.TOOL_CALLING_LLM);
builder.addConditionalEdges(
  NodeNames.TOOL_CALLING_LLM,
  toolsCondition
);
builder.addEdge(NodeNames.TOOLS, END);

// Compile graph
export const graph = builder.compile();

Error Message and Stack Trace (if applicable)

When I try to run the graph in LangGraph studio, the tools node always fail:

{
  "messages": [
    {
      "status": "error",
      "content": "Error: Received tool input did not match expected schema\n Please fix your mistakes.",
      "name": "multiply",
      "tool_call_id": "toolu_01Mdkq5hFSaAUmAbwruGUJ34",
      "additional_kwargs": {},
      "response_metadata": {},
      "type": "tool"
    },
    {
      "status": "error",
      "content": "Error: Tool \"\" not found.\n Please fix your mistakes.",
      "name": "",
      "tool_call_id": "fallback-1",
      "additional_kwargs": {},
      "response_metadata": {},
      "type": "tool"
    }
  ]
}

Description

Hello, I use LangGraph to build a simple router graph with tool calling.

TypeScript code for the graph router.ts is as follows:

// Tool
const multiplyTool = tool(
  async ({ a, b }: { a: number; b: number }) => {
    console.log(`Executing multiply tool: ${a}*${b}`);
    return a * b;
  },
  {
    name: "multiply",
    description: "Multiplies a and b.",
    schema: z.object({
      a: z.number().describe("first number"),
      b: z.number().describe("second number"),
    }),
  }
);

// LLM with bound tool
const llm = new ChatAnthropic({ model: "claude-3-5-haiku-latest", temperature: 0 });
const tools = [multiplyTool];
const llmWithTools = llm.bindTools(tools);

// Node
async function toolCallingLLM(state: typeof MessagesAnnotation.State) {
  const result = await llmWithTools.invoke(state.messages);

  return { messages: [result] };
}

enum NodeNames {
  TOOL_CALLING_LLM = "tool_calling_llm",
  TOOLS = "tools"
}

// Build graph
const builder = new StateGraph<typeof MessagesAnnotation.spec, any, any, NodeNames>(MessagesAnnotation);
builder.addNode(NodeNames.TOOL_CALLING_LLM, toolCallingLLM);
builder.addNode(NodeNames.TOOLS, new ToolNode(tools));
builder.addEdge(START, NodeNames.TOOL_CALLING_LLM);
builder.addConditionalEdges(
  NodeNames.TOOL_CALLING_LLM,
  toolsCondition
);
builder.addEdge(NodeNames.TOOLS, END);

// Compile graph
export const graph = builder.compile();

Everything works fine when I simply invoke the graph:

const messages = await graph.invoke({
    messages: [new HumanMessage({ content: "What is 6 multiplied by 7?" })]
  });

The tool_calling_llm node will respond with AIMessage where the tool call for multiply tool is in one single tool call:

AIMessage {
  "id": "msg_01DkVozCzrgPJPRCfMTLEphQ",
  "content": [
    {
      "type": "text",
      "text": "I'll help you calculate that by using the multiply function."
    },
    {
      "type": "tool_use",
      "id": "toolu_0115uXLnXQ3iFXXky4HYRAFj",
      "name": "multiply",
      "input": {
        "a": 6,
        "b": 7
      }
    }
  ],
  ...
  "tool_calls": [
    {
      "name": "multiply",
      "args": {
        "a": 6,
        "b": 7
      },
      "id": "toolu_0115uXLnXQ3iFXXky4HYRAFj",
      "type": "tool_call"
    }
  ],
  ...
}

However, when I try to run the graph in LangGraph studio, the tools node always fail:

{
  "messages": [
    {
      "status": "error",
      "content": "Error: Received tool input did not match expected schema\n Please fix your mistakes.",
      "name": "multiply",
      "tool_call_id": "toolu_01Mdkq5hFSaAUmAbwruGUJ34",
      "additional_kwargs": {},
      "response_metadata": {},
      "type": "tool"
    },
    {
      "status": "error",
      "content": "Error: Tool \"\" not found.\n Please fix your mistakes.",
      "name": "",
      "tool_call_id": "fallback-1",
      "additional_kwargs": {},
      "response_metadata": {},
      "type": "tool"
    }
  ]
}

This is because tool_calling_llm node responds with AIMessageChunk.
The tool call is split into 2 calls, one for the tool name and the other for the input:

AIMessageChunk {
  "id": "msg_01JzyTs4eo3XW6owEBCSEhMe",
  "content": [
    {
      "index": 0,
      "type": "text",
      "text": "I'll help you calculate 1 times 2 using the multiply function."
    },
    {
      "index": 1,
      "type": "tool_use",
      "id": "toolu_01Mdkq5hFSaAUmAbwruGUJ34",
      "name": "multiply",
      "input": ""
    },
    {
      "index": 1,
      "input": "{\"a\": 1, \"b\": 2}",
      "type": "input_json_delta"
    }
  ],
 
  ....

  "tool_calls": [
    {
      "name": "multiply",
      "args": {},
      "id": "toolu_01Mdkq5hFSaAUmAbwruGUJ34",
      "type": "tool_call"
    },
    {
      "name": "",
      "args": {
        "a": 1,
        "b": 2
      },
      "id": "fallback-1",
      "type": "tool_call"
    }
  ],
  "tool_call_chunks": [
    {
      "id": "toolu_01Mdkq5hFSaAUmAbwruGUJ34",
      "index": 1,
      "name": "multiply",
      "args": ""
    },
    {
      "index": 1,
      "args": "{\"a\": 1, \"b\": 2}"
    }
  ],
  ....
}

I have tried disabling messages stream mode next to the Submit button. The issue still persists.

I could reproduce the issue outside LangGraph Studio, if I invoke the graph with stream mode "messages""

 const messages = await graph.invoke({
    messages: [new HumanMessage({ content: "What is 6 multiplied by 7?" })]
  }, { streamMode: ["values", "messages"] });

It looks like disabling messages streaming in LangGraph Studio no longer works.
This prevents me from debugging/testing my graph in LangGraph Studio.
I am using LangGraph cli to run my graph. So there is an issue with LangGraph.js
Appreciate if you could look into this issue.

langgraphjs dev

System Info

"dependencies": {
"@langchain/anthropic": "^0.3.0",
"@langchain/core": "^0.3.76",
"@langchain/langgraph": "^0.4.9",
"dotenv": "^16.0.0",
"zod": "^3.22.0"
}