[Request] 建议mcp服务调用支持提示词模式

[chung1912]

🥰 需求描述

目前mcp服务通过工具调用的模式使用，然而，有些模型没有工具调用的能力（例如阿里的deepseek-v3），有的情况是供应商的工具调用规则比较另类，使用的时候各种报错，无法使用mcp

目前像cherry，cline等项目都是采用提示词的模式调用工具，使用效果也较好，可以兼容所有的模型

🧐 解决方案

通过后端环境变量配置，默认为false，保持为现行的工具调用模式使用mcp，当设置为true时，则使用提示词的方式使用mcp服务

📝 补充信息

No response

[lobehubbot]

👀 @chung1912

Thank you for raising an issue. We will investigate into the matter and get back to you as soon as possible.
Please make sure you have given us as much context as possible.
非常感谢您提交 issue。我们会尽快调查此事，并尽快回复您。 请确保您已经提供了尽可能多的背景信息。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

🥰 Requirement description

Currently, mcp service is used through tool calling mode. However, some models do not have the ability to call tool (such as Alibaba's deepseek-v3). In some cases, the supplier's tool calling rules are relatively alternative, and various errors are reported when using it, so mcp cannot be used.

Currently, projects such as cherry and cline use pattern calling tools that use prompt words, and the usage effect is also good and can be compatible with all models

🧐 Solution

Configure the backend environment variables, default to false, keep mcp in the current tool calling mode, when set to true, use mcp service using prompt words

📝 Supplementary information

No response

[arvinxx]

不考虑。这种方式会导致很多问题：

1. 平行 tools calling 的流式不可能实现；
2. 没法做到单个 tool 的结构化解析和展示，并且有很大概率会有生成出错的问题（Cherry Studio 被很多人诟病的 MCP 调用不稳定就是这样带来的，但其实是模型能力不够强）；
3. 没法借助模型能力提升解锁更多场景，这和行业趋势的迭代背道而驰了；
4. 我们的 DB 结构是专门针对 plugin（Tools Calling）做了适配，用提示词方案就没法把这种结构化数据利用起来，属于放弃了我们原有的优势

[lobehubbot]

✅ @chung1912

This issue is closed, If you have any questions, you can comment and reply.
此问题已经关闭。如果您有任何问题，可以留言并回复。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Not considered. This method can lead to many problems:

1. The streaming method of parallel tools calling is impossible;
2. It is impossible to perform structured analysis and display of a single tool, and there is a high probability that there will be generation errors (Cherry Studio is criticized by many people for its unstable MCP call, which is caused by this, but in fact it is because the model capability is not strong enough);
3. It is impossible to unlock more scenarios with the help of model capabilities, which runs contrary to the iteration of industry trends;
4. Our DB structure is specifically adapted to plugin (Tools Calling). This structured data cannot be utilized with the prompt word scheme, which is absent from our original advantages.