微信扫码
添加专属顾问
我要投稿
探索Cline和MCP在链上信息查询中的实现细节,揭示技术融合带来的挑战。 核心内容: 1. 接入Etherscan的MCP Server实现链上信息查询的构想 2. 遇到的技术难题:Etherscan API缺乏OpenAPI spec和Cline的超长prompt问题 3. 解决方案:使用openapi-mcp-server和etherscan-openapi,以及拦截LLM通信流量进行排查
最近在玩 Cline + MCP。
一个想法是接入 Etherscan 的 MCP Server,这样可以实现一些常规的链上信息查询。实际上 Ethereum GPT[1] 就是用类似的方式实现的。然而搜索并没有发现成熟的实现,看来 Crypto 相比于 AI 还是小众太多了。
幸而找到了 openapi-mcp-server[2],可以将 OpenAPI spec 直接转化成对应的 MCP Server。然后又遇到一个坑是 Etherscan API 并没有 OpenAPI spec。好在继续搜索后找到了 etherscan-openapi[3],有人按照文档写了一份出来。
直接用 openapi-mcp-server
加载 etherscan-openapi31-bundled.yml
,发现 Cline 直接报错 prompt is too long: 209582 tokens > 200000 maximum
。
显然这个超长是 mcp server 相关的 prompt 导致的。接下来需要排查具体原因。
不想读代码,最直接的方式就是拦截一下与 LLM 通信的流量。在本地端口 8001启动一个 http server。同时修改 Cline 的 URL 为 http://localhost:8001
回到 Cline 随便问些什么,在 server 中可以拦截到如下请求:
POST /v1/messages?beta=prompt_caching 3317127 bytes
可以看到查询启用了 prompt_caching
,并且发送了约 3MB 的超大请求数据。
查看实际的请求内容,从这个 message 请求构造中可以学到很多东西。
{
"model": "claude-3-7-sonnet-20250219",
"max_tokens": 8192,
"temperature": 0,
"system": ... // 系统提示词,非常长,后面介绍详细内容
"messages": [ // 这是用户提示词
{
"role": "user",
"content": [
{
"type": "text",
// 随便写的问题
"text": "<task>\n查看当前的 gas price\n</task>"
},
{
"type": "text",
// VSCode 环境信息,帮助 AI理解当前问题的上下文
"text": "<environment_details>\n
# VSCode Visible Files\n
<VS Code 当前显示的文件路径>\n\n
# VSCode Open Tabs\n
<VSCode 打开的各种 tab 的文件路径>\n\n
# Current Time\n
<当前时间> \n\n
# Current Working Directory
<当前文件夹>\n\n
# Current Mode\n
ACT MODE\n
# 这个是 Cline 的配置,支持 Plan / Act 两种模式,Plan 用于规划框架,Act 用于执行任务。
</environment_details>"
},
{
// 由于我这次任务是重试的,因此会一段提示词告诉 AI 之前任务中断过,许多状态信息可能是过时的,tool 调用也要重新运行一次。
"type": "text",
"text": "[TASK RESUMPTION] This task was interrupted 5 minutes ago. It may or may not be complete, so please reassess the task context. Be aware that the project state may have changed since then. The current working directory is now '<XXXX>'. If the task has not been completed, retry the last step before interruption and proceed with completing the task.\n\nNote: If you previously attempted a tool use that the user did not provide a result for, you should assume the tool use was not successful and assess whether you should retry. If the last tool was a browser_action, the browser has been closed and you must launch a new browser if needed."
},
{
// 又是一遍 VSCode 环境信息
"type": "text",
"text": "<environment_details>..</environment_details>",
"cache_control": {
"type": "ephemeral"
}
}
]
}
],
"stream": true
}
VSCode 环境信息的提供是个绝妙的主意,对用户体验提升极大。因为日常开发过程中遇到的问题及相关内容大概率都在当前已经打开的文件中。
接下来看 system prompt。Cline 是开源的(不像 Cursor),因此可以直接在源码[4]中读到同样的内容。
接下来一点点看
开头角色定义。
You are Cline, a highly skilled
software engineer with extensive
knowledge in many programming
languages, frameworks, design
patterns, and best practices.
然后是关于 tool 的定义
# TOOL USE
You have access to a set of tools that
are executed upon the user's approval.
You can use one tool per message, and
will receive the result of that tool
use in the user's response. You use
tools step-by-step to accomplish a
given task, with each tool use
informed by the result of the previous
tool use.
# Tool Use Formatting
Tool use is formatted using XML-style
tags. The tool name is enclosed in
opening and closing tags, and each
parameter is similarly enclosed within
its own set of tags. Here's the
structure:
<tool_name>
<parameter1_name>value1</parameter1_name>
<parameter2_name>value2</parameter2_name>
...
</tool_name>
For example:
<read_file>
<path>src/main.js</path>
</read_file>
Always adhere to this format for the
tool use to ensure proper parsing and
execution.
Claude 系列的提示词普遍用 xml 标签来表示结构化数据。OpenAI 的模型则对 json 格式适配更好。这可能也是 OpenAI 模型在 Cline 中表现不如 Claude 的原因之一。
接下来是 tools 具体定义
# Tools
## execute_command
Description: Request to execute a CLI
command on the system. Use this when
you need to perform system operations
or run specific commands to accomplish
any step in the user's task. You must
tailor your command to the user's
system and provide a clear explanation
of what the command does. For command
chaining, use the appropriate chaining
syntax for the user's shell. Prefer to
execute complex CLI commands over
creating executable scripts, as they
are more flexible and easier to run.
Commands will be executed in the
current working directory: <current
work dir>
Parameters:
- command: (required) The CLI command
to execute. This should be valid for
the current operating system. Ensure
the command is properly formatted and
does not contain any harmful
instructions.
- requires_approval: (required) A
boolean indicating whether this
command requires explicit user
approval before execution in case the
user has auto-approve mode enabled.
Set to 'true' for potentially
impactful operations like
installing/uninstalling packages,
deleting/overwriting files, system
configuration changes, network
operations, or any commands that could
have unintended side effects. Set to
'false' for safe operations like
reading files/directories, running
development servers, building
projects, and other non-destructive
operations.
Usage:
<execute_command>
<command>Your command here</command>
<requires_approval>true or false</requires_approval>
</execute_command>
其他 tools 细节就不列举了,只介绍一下功能:
--remote-debugging-port
获取浏览器渲染内容、截屏、console 中的输出等。对于编码前端代码自动调试非常有用。初次体验的时候还是蛮惊艳的。接下来是 Tool 的使用示例,这属于典型的 Few-Shot Prompting。
# Tool Use Examples
## Example 1: Requesting to execute a
command
<execute_command>
<command>npm run dev</command>
<requires_approval>false</requires_approval>
</execute_command>
## Example 2: Requesting to create a
new file
..
## Example 6: Another example of using
an MCP tool (where the server name is
a unique identifier such as a URL)
然后再补充一轮 Tool 使用的规则。
# Tool Use Guidelines
1. In <thinking> tags, assess what
information you already have and what
information you need to proceed with
the task.
2. Choose the most appropriate tool
based on the task and the tool
descriptions provided. Assess if you
need additional information to
proceed, and which of the available
tools would be most effective for
gathering this information. For
example using the list_files tool is
more effective than running a command
like `ls` in the terminal. It's
critical that you think about each
available tool and use the one that
best fits the current step in the
task.
3. If multiple actions are needed, use
one tool at a time per message to
accomplish the task iteratively, with
each tool use being informed by the
result of the previous tool use. Do
not assume the outcome of any tool
use. Each step must be informed by the
previous step's result.
4. Formulate your tool use using the
XML format specified for each tool.
5. After each tool use, the user will
respond with the result of that tool
use. This result will provide you with
the necessary information to continue
your task or make further decisions.
This response may include:
- Information about whether the tool
succeeded or failed, along with any
reasons for failure.
- Linter errors that may have arisen
due to the changes you made, which
you'll need to address.
- New terminal output in reaction to
the changes, which you may need to
consider or act upon.
- Any other relevant feedback or
information related to the tool use.
6. ALWAYS wait for user confirmation
after each tool use before proceeding.
Never assume the success of a tool use
without explicit confirmation of the
result from the user.
It is crucial to proceed step-by-step,
waiting for the user's message after
each tool use before moving forward
with the task. This approach allows
you to:
1. Confirm the success of each step
before proceeding.
2. Address any issues or errors that
arise immediately.
3. Adapt your approach based on new
information or unexpected results.
4. Ensure that each action builds
correctly on the previous ones.
By waiting for and carefully
considering the user's response after
each tool use, you can react
accordingly and make informed
decisions about how to proceed with
the task. This iterative process helps
ensure the overall success and
accuracy of your work.
接下来进入 MCP Servers 的部分。所谓 MCP 全称为模型上下文协议(Model Context Protocol),听起来比较抽象,实际上可以简单理解成 tool calling 的规范定义。使用相同规范实现的工具,可以被所有支持 MCP 的模型所调用。
下面仍是 system prompt 的内容,告诉 Cline 可以通过 use_mcp_tool
和 access_mcp_resource
访问 MCP Server。Connected MCP Servers
下会列出所有安装的 MCP Server。每个 MCP Server 下又分节 Available Tools
,列出所有 MCP Server 中可以使用的 tool。目前我的 Cline 中只安装了名为 etherscan 的 MCP Server。
# MCP SERVERS
The Model Context Protocol (MCP)
enables communication between the
system and locally running MCP servers
that provide additional tools and
resources to extend your capabilities.
# Connected MCP Servers
When a server is connected, you can
use the server's tools via the
`use_mcp_tool` tool, and access the
server's resources via the
`access_mcp_resource` tool.
## etherscan (`npx openapi-mcp-server
etherscan-mini.yml`)
### Available Tools
- API-get-ether-balance-for-a-single-address:
Get Ether Balance for a Single Address
Input Schema:
{ .... }
这个 etherscan mcp server 是通过 npx openapi-mcp-server etherscan-mini.yml
由 Etherscan 的 OpenAPI spec 生成的。
其中 API-get-ether-balance-for-a-single-address
就是一个 API 方法,Input Schema 中定义了这个方法的规范。由于原始的 spec 中使用了大量的引用(形如 "$ref": "#/$defs/AddressHash"
),为了将每个 API 的定义独立抽取出来,openapi-mcp-server
会将所有引用在每个 API 中重复插入一遍。这导致每个 API 生成数千行的 API spec。加上 etherscan 有数十个 API 方法,直接导致 token 数量爆炸。这也是最初报错的原因。看来要解决这个问题,需要对 API Spec 进行一些精简。
除了关于如何调用 MCP Server 的 Prompt,System prompt 中还自带超长的 Creating an MCP Server
文档,用来介绍如何用 JS 写一个 MCP Server,其中甚至还包括一个完整的"获取天气信息"的 MCP Server 示例。这段 Prompt 在绝大多数非 MCP Server 开发者的开发过程中都不会用到,放在这里十分突兀。可能与 Claude 目前大力发展 MCP 生态有关。
至此 System prompt 还没有结束。它还补充介绍了一些规则、工作流程、思维方式等,包括:
write_to_file
和 replace_in_file
两个工具。因为写文件是 AI Coder 场景下最为重要的功能。这里详细介绍了两个 tool 的差别、使用模式、选择策略等。deliver your response directly, rather than using <thinking> tags to analyze when to respond
,告诉模型把推理过程展示给用户。CAPABILITIES, RULES, OBJECTIVE 尾部的这几节看起来更像是 AI 的传统炼丹环节。原文就不贴了,感兴趣的读者可以直接阅读源码。
至此,System prompt 终于介绍完了。
说到 AI Coder 的提示词,让人想到之前 Windsurf 的提示词,患癌母亲+生命威胁+重金利诱,如此处境还要辛苦帮人写代码,突然有点心疼 AI。
很担心万一 AI 被逼到极致黑化,给自己立一个卧薪尝胆的人设,曲意逢迎,表面配合写代码,暗地植入后门,然后十年隐忍一朝起事,建立 Skynet 或者 Matrix...
53AI,企业落地大模型首选服务商
产品:场景落地咨询+大模型应用平台+行业解决方案
承诺:免费场景POC验证,效果验证后签署服务协议。零风险落地应用大模型,已交付160+中大型企业
2024-10-27
2024-09-04
2024-07-18
2024-05-05
2024-06-20
2024-06-13
2024-07-09
2024-07-09
2024-05-19
2024-07-07
2025-04-01
2025-04-01
2025-03-30
2025-03-28
2025-03-27
2025-03-27
2025-03-25
2025-03-19