Hacktoberfest 2026 : les issues que les mainteneurs ont marquées pour octobre, ouvertes et accessibles aux débutants. Parcourir les issues Hacktoberfest

Bug in the official tutorial docs: Tool use example code lacks agentic loop, causing multi-turn calls to fail

Ouverte
  • Quelque chose ne va pas ?
#233 1 commentaire 2 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 6 jours

Personne n'a encore pris cette issue.

Évaluation

Difficulté
3/5
Temps estimé
1-2 jours
Accessibilité débutants
76/100
Type d'issue
Bug
Clarté
Clairement spécifiée
Activité
Active
Stack technique
python
Domaine
documentation

Piste de recherche

Start with “The Query Processing Logic” in the MCP building tutorial at the linked build-client page, focusing on the async process_query function and its message handling. Confirm that repeated tool_use responses are processed with their tool results until a plain-text response is returned, including tool execution errors; the issue’s corrected snippet defines the expected behavior.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

The original example code (The Query Processing Logic in the MCP building tutorial site https://modelcontextprotocol.io/docs/2026-07-28/develop/build-client) has the following core issues:

  • Single-turn Limitation: After the initial API call, the code only checks and processes tool_use blocks once. If Claude requests additional tool calls in its second response (e.g., needing to chain a second tool based on the output of the first), the original code fails to handle them—it simply ignores subsequent tool calls, causing complex multi-step tasks to fail.
  • Lack of Agentic Loop Structure: It does not use a loop to continuously maintain conversation state. A correct implementation should, after each model response, check whether tool_use blocks are present; if so, execute the tools, append the results to the message history, and continue with the next API call—repeating this cycle until the model outputs a plain-text final answer.

The following corrected snippet resolves these issues by introducing a while True loop, correctly implementing a multi-turn tool-calling agent loop. Below is the corrected code version:

async def process_query(client, query: str) -> str:
    
    messages = [
        {"role": "user",
         "content": query}
    ]

    tool_list = await client.list_tools()
    available_tools = [{
        "name": tool.name,
        "description": tool.description,
        "input_schema": tool.input_schema,
    } for tool in tool_list.tools]

    final_texts = []

    while True:
        response = await anthropic.messages.create(
            model=MODEL,
            max_tokens=1000,
            messages=messages,
            tools=available_tools,
        )

        # Append the full assistant response back to the message history for context
        messages.append({
            "role": "assistant",
            "content": response.content,
        })

        tool_uses = []

        for block in response.content:
            if block.type == "text":
                final_texts.append(block.text)
            elif block.type == "tool_use":
                tool_uses.append(block)

        # If no tool_use is requested, this is the final answer; exit the loop
        if not tool_uses:
            break

        tool_results = []

        for block in tool_uses:
            try:
                result = await client.call_tool(block.name, block.input)

                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": format_tool_result(result),  # Assumes format_tool_result is defined elsewhere
                    "is_error": bool(result.is_error),
                })
            except Exception as e:
                # Tool execution errors must also be returned as tool_result to the model
                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": f"Tool execution error: {e}",
                    "is_error": True,
                })

        # Feed the tool results back as a user message to continue the next turn
        messages.append({
            "role": "user",
            "content": tool_results,
        })

    return "\n".join(final_texts)

(Note: format_tool_result is assumed to be a helper function that extracts text from the tool response, similar to "\n".join(block.text for block in result.content if isinstance(block, TextContent)) in the original code.)

Langage dominant
TypeScript
Étoiles
1.2k
Forks
693
Merge moyen
4 j 1 h
PR mergées (30 j)
27

Préparer son environnement

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de modelcontextprotocol/quickstart-resources

Toutes les issues de modelcontextprotocol/quickstart-resources

Issues similaires

Plus d'issues TypeScript

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.