mirror of
https://github.com/CherryHQ/cherry-studio.git
synced 2025-12-27 21:01:32 +08:00
feat: enhance MCP Prompt plugin and recursive call capabilities
- Updated `tsconfig.web.json` to support wildcard imports for `@cherrystudio/ai-core`. - Enhanced `package.json` to include type definitions and imports for built-in plugins. - Introduced recursive call functionality in `PluginManager` and `PluginEngine`, allowing for improved handling of tool interactions. - Added `MCPPromptPlugin` to facilitate tool calls within prompts, enabling recursive processing of tool results. - Refactored `transformStream` methods across plugins to accommodate new parameters and improve type safety.
This commit is contained in:
MyPrototypeWhat
parent
8b67a45804
commit
9293f26612
@ -118,6 +118,11 @@
|
||||
"types": "./src/index.ts",
|
||||
"import": "./src/index.ts",
|
||||
"require": "./src/index.ts"
|
||||
},
|
||||
"./core/plugins/built-in": {
|
||||
"types": "./src/core/plugins/built-in/index.ts",
|
||||
"import": "./src/core/plugins/built-in/index.ts",
|
||||
"require": "./src/core/plugins/built-in/index.ts"
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@ -5,3 +5,5 @@
|
||||
export const BUILT_IN_PLUGIN_PREFIX = 'built-in:'
|
||||
|
||||
export { createLoggingPlugin } from './logging'
|
||||
export type { MCPPromptConfig, ToolUseResult } from './mcpPrompt'
|
||||
export { createMCPPromptPlugin } from './mcpPrompt'
|
||||
|
||||
670
packages/aiCore/src/core/plugins/built-in/mcpPrompt.ts
Normal file
670
packages/aiCore/src/core/plugins/built-in/mcpPrompt.ts
Normal file
@ -0,0 +1,670 @@
|
||||
/**
|
||||
* 内置插件：MCP Prompt 模式
|
||||
* 为不支持原生 Function Call 的模型提供 prompt 方式的工具调用
|
||||
* 内置默认逻辑，支持自定义覆盖
|
||||
*/
|
||||
import { ToolSet } from 'ai'
|
||||
|
||||
import { definePlugin } from '../index'
|
||||
import type { AiRequestContext } from '../types'
|
||||
|
||||
/**
|
||||
* 使用 AI SDK 的 Tool 类型，更通用
|
||||
*/
|
||||
// export interface Tool {
|
||||
// type: 'function'
|
||||
// function: {
|
||||
// name: string
|
||||
// description?: string
|
||||
// parameters?: {
|
||||
// type: 'object'
|
||||
// properties: Record<string, any>
|
||||
// required?: string[]
|
||||
// additionalProperties?: boolean
|
||||
// }
|
||||
// }
|
||||
// }
|
||||
|
||||
/**
|
||||
* 解析结果类型
|
||||
* 表示从AI响应中解析出的工具使用意图
|
||||
*/
|
||||
export interface ToolUseResult {
|
||||
id: string
|
||||
toolName: string
|
||||
arguments: any
|
||||
status: 'pending' | 'invoking' | 'done' | 'error'
|
||||
}
|
||||
|
||||
/**
|
||||
* MCP Prompt 插件配置
|
||||
*/
|
||||
export interface MCPPromptConfig {
|
||||
// 是否启用（用于运行时开关）
|
||||
enabled?: boolean
|
||||
// 自定义系统提示符构建函数（可选，有默认实现）
|
||||
buildSystemPrompt?: (userSystemPrompt: string, tools: ToolSet) => Promise<string>
|
||||
// 自定义工具解析函数（可选，有默认实现）
|
||||
parseToolUse?: (content: string, tools: ToolSet) => ToolUseResult[]
|
||||
}
|
||||
|
||||
// 全局存储，解决 transformStream 中无 context 的问题
|
||||
const globalToolsStorage = new Map<string, ToolSet>()
|
||||
|
||||
/**
|
||||
* 生成唯一的执行ID
|
||||
*/
|
||||
function generateExecutionId(): string {
|
||||
return `mcp_${Date.now()}_${Math.random().toString(36).slice(2)}`
|
||||
}
|
||||
|
||||
/**
|
||||
* 存储工具信息
|
||||
*/
|
||||
|
||||
/**
|
||||
* 全局存储工具信息
|
||||
*/
|
||||
function storeGlobalTools(executionId: string, tools: ToolSet) {
|
||||
globalToolsStorage.set(executionId, tools)
|
||||
}
|
||||
|
||||
/**
|
||||
* 获取全局存储的工具信息
|
||||
*/
|
||||
function getGlobalTools(executionId: string): ToolSet | undefined {
|
||||
return globalToolsStorage.get(executionId)
|
||||
}
|
||||
|
||||
/**
|
||||
* 清理全局存储
|
||||
*/
|
||||
function clearGlobalTools(executionId: string) {
|
||||
globalToolsStorage.delete(executionId)
|
||||
}
|
||||
|
||||
/**
|
||||
* 默认系统提示符模板（提取自 Cherry Studio）
|
||||
*/
|
||||
const DEFAULT_SYSTEM_PROMPT = `In this environment you have access to a set of tools you can use to answer the user's question. \\
|
||||
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_use>
|
||||
<name>{tool_name}</name>
|
||||
<arguments>{json_arguments}</arguments>
|
||||
</tool_use>
|
||||
|
||||
The tool name should be the exact name of the tool you are using, and the arguments should be a JSON object containing the parameters required by that tool. For example:
|
||||
<tool_use>
|
||||
<name>python_interpreter</name>
|
||||
<arguments>{"code": "5 + 3 + 1294.678"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
The user will respond with the result of the tool use, which should be formatted as follows:
|
||||
|
||||
<tool_use_result>
|
||||
<name>{tool_name}</name>
|
||||
<result>{result}</result>
|
||||
</tool_use_result>
|
||||
|
||||
The result should be a string, which can represent a file or any other output type. You can use this result as input for the next action.
|
||||
For example, if the result of the tool use is an image file, you can use it in the next action like this:
|
||||
|
||||
<tool_use>
|
||||
<name>image_transformer</name>
|
||||
<arguments>{"image": "image_1.jpg"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
Always adhere to this format for the tool use to ensure proper parsing and execution.
|
||||
|
||||
## Tool Use Examples
|
||||
{{ TOOL_USE_EXAMPLES }}
|
||||
|
||||
## Tool Use Available Tools
|
||||
Above example were using notional tools that might not exist for you. You only have access to these tools:
|
||||
{{ AVAILABLE_TOOLS }}
|
||||
|
||||
## Tool Use Rules
|
||||
Here are the rules you should always follow to solve your task:
|
||||
1. Always use the right arguments for the tools. Never use variable names as the action arguments, use the value instead.
|
||||
2. Call a tool only when needed: do not call the search agent if you do not need information, try to solve the task yourself.
|
||||
3. If no tool call is needed, just answer the question directly.
|
||||
4. Never re-do a tool call that you previously did with the exact same parameters.
|
||||
5. For tool use, MAKE SURE use XML tag format as shown in the examples above. Do not use any other format.
|
||||
|
||||
# User Instructions
|
||||
{{ USER_SYSTEM_PROMPT }}
|
||||
|
||||
Now Begin! If you solve the task correctly, you will receive a reward of $1,000,000.`
|
||||
|
||||
/**
|
||||
* 默认工具使用示例（提取自 Cherry Studio）
|
||||
*/
|
||||
const DEFAULT_TOOL_USE_EXAMPLES = `
|
||||
Here are a few examples using notional tools:
|
||||
---
|
||||
User: Generate an image of the oldest person in this document.
|
||||
|
||||
A: I can use the document_qa tool to find out who the oldest person is in the document.
|
||||
<tool_use>
|
||||
<name>document_qa</name>
|
||||
<arguments>{"document": "document.pdf", "question": "Who is the oldest person mentioned?"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
User: <tool_use_result>
|
||||
<name>document_qa</name>
|
||||
<result>John Doe, a 55 year old lumberjack living in Newfoundland.</result>
|
||||
</tool_use_result>
|
||||
|
||||
A: I can use the image_generator tool to create a portrait of John Doe.
|
||||
<tool_use>
|
||||
<name>image_generator</name>
|
||||
<arguments>{"prompt": "A portrait of John Doe, a 55-year-old man living in Canada."}</arguments>
|
||||
</tool_use>
|
||||
|
||||
User: <tool_use_result>
|
||||
<name>image_generator</name>
|
||||
<result>image.png</result>
|
||||
</tool_use_result>
|
||||
|
||||
A: the image is generated as image.png
|
||||
|
||||
---
|
||||
User: "What is the result of the following operation: 5 + 3 + 1294.678?"
|
||||
|
||||
A: I can use the python_interpreter tool to calculate the result of the operation.
|
||||
<tool_use>
|
||||
<name>python_interpreter</name>
|
||||
<arguments>{"code": "5 + 3 + 1294.678"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
User: <tool_use_result>
|
||||
<name>python_interpreter</name>
|
||||
<result>1302.678</result>
|
||||
</tool_use_result>
|
||||
|
||||
A: The result of the operation is 1302.678.
|
||||
|
||||
---
|
||||
User: "Which city has the highest population , Guangzhou or Shanghai?"
|
||||
|
||||
A: I can use the search tool to find the population of Guangzhou.
|
||||
<tool_use>
|
||||
<name>search</name>
|
||||
<arguments>{"query": "Population Guangzhou"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
User: <tool_use_result>
|
||||
<name>search</name>
|
||||
<result>Guangzhou has a population of 15 million inhabitants as of 2021.</result>
|
||||
</tool_use_result>
|
||||
|
||||
A: I can use the search tool to find the population of Shanghai.
|
||||
<tool_use>
|
||||
<name>search</name>
|
||||
<arguments>{"query": "Population Shanghai"}</arguments>
|
||||
</tool_use>
|
||||
|
||||
User: <tool_use_result>
|
||||
<name>search</name>
|
||||
<result>26 million (2019)</result>
|
||||
</tool_use_result>
|
||||
Assistant: The population of Shanghai is 26 million, while Guangzhou has a population of 15 million. Therefore, Shanghai has the highest population.`
|
||||
|
||||
/**
|
||||
* 构建可用工具部分（提取自 Cherry Studio）
|
||||
*/
|
||||
function buildAvailableTools(tools: ToolSet): string {
|
||||
const availableTools = Object.keys(tools)
|
||||
.map((toolName: string) => {
|
||||
const tool = tools[toolName]
|
||||
return `
|
||||
<tool>
|
||||
<name>${toolName}</name>
|
||||
<description>${tool.description || ''}</description>
|
||||
<arguments>
|
||||
${tool.parameters ? JSON.stringify(tool.parameters) : ''}
|
||||
</arguments>
|
||||
</tool>
|
||||
`
|
||||
})
|
||||
.join('\n')
|
||||
return `<tools>
|
||||
${availableTools}
|
||||
</tools>`
|
||||
}
|
||||
|
||||
/**
|
||||
* 默认的系统提示符构建函数（提取自 Cherry Studio）
|
||||
*/
|
||||
async function defaultBuildSystemPrompt(userSystemPrompt: string, tools: ToolSet): Promise<string> {
|
||||
const availableTools = buildAvailableTools(tools)
|
||||
|
||||
const fullPrompt = DEFAULT_SYSTEM_PROMPT.replace('{{ TOOL_USE_EXAMPLES }}', DEFAULT_TOOL_USE_EXAMPLES)
|
||||
.replace('{{ AVAILABLE_TOOLS }}', availableTools)
|
||||
.replace('{{ USER_SYSTEM_PROMPT }}', userSystemPrompt || '')
|
||||
|
||||
return fullPrompt
|
||||
}
|
||||
|
||||
/**
|
||||
* 默认工具解析函数（提取自 Cherry Studio）
|
||||
* 解析 XML 格式的工具调用
|
||||
*/
|
||||
function defaultParseToolUse(content: string, tools: ToolSet): ToolUseResult[] {
|
||||
if (!content || !tools || Object.keys(tools).length === 0) {
|
||||
return []
|
||||
}
|
||||
|
||||
// 支持两种格式：
|
||||
// 1. 完整的 <tool_use></tool_use> 标签包围的内容
|
||||
// 2. 只有内部内容（从 TagExtractor 提取出来的）
|
||||
|
||||
let contentToProcess = content
|
||||
|
||||
// 如果内容不包含 <tool_use> 标签，说明是从 TagExtractor 提取的内部内容，需要包装
|
||||
if (!content.includes('<tool_use>')) {
|
||||
contentToProcess = `<tool_use>\n${content}\n</tool_use>`
|
||||
}
|
||||
|
||||
const toolUsePattern =
|
||||
/<tool_use>([\s\S]*?)<name>([\s\S]*?)<\/name>([\s\S]*?)<arguments>([\s\S]*?)<\/arguments>([\s\S]*?)<\/tool_use>/g
|
||||
const results: ToolUseResult[] = []
|
||||
let match
|
||||
let idx = 0
|
||||
|
||||
// Find all tool use blocks
|
||||
while ((match = toolUsePattern.exec(contentToProcess)) !== null) {
|
||||
const toolName = match[2].trim()
|
||||
const toolArgs = match[4].trim()
|
||||
|
||||
// Try to parse the arguments as JSON
|
||||
let parsedArgs
|
||||
try {
|
||||
parsedArgs = JSON.parse(toolArgs)
|
||||
} catch (error) {
|
||||
// If parsing fails, use the string as is
|
||||
parsedArgs = toolArgs
|
||||
}
|
||||
|
||||
// Find the corresponding tool
|
||||
const tool = tools[toolName]
|
||||
if (!tool) {
|
||||
console.warn(`Tool "${toolName}" not found in available tools`)
|
||||
continue
|
||||
}
|
||||
|
||||
// Add to results array
|
||||
results.push({
|
||||
id: `${toolName}-${idx++}`, // Unique ID for each tool use
|
||||
toolName: toolName,
|
||||
arguments: parsedArgs,
|
||||
status: 'pending'
|
||||
})
|
||||
}
|
||||
return results
|
||||
}
|
||||
|
||||
/**
|
||||
* 创建 MCP Prompt 插件
|
||||
*/
|
||||
export const createMCPPromptPlugin = definePlugin((config: MCPPromptConfig = {}) => {
|
||||
const { enabled = true, buildSystemPrompt = defaultBuildSystemPrompt, parseToolUse = defaultParseToolUse } = config
|
||||
|
||||
// 为每个插件实例生成唯一ID
|
||||
const executionId = generateExecutionId()
|
||||
|
||||
return {
|
||||
name: 'built-in:mcp-prompt',
|
||||
|
||||
transformParams: async (params: any, context: AiRequestContext) => {
|
||||
if (!enabled || !params.tools) return params
|
||||
|
||||
// 保存原始工具信息到 WeakMap 和全局存储中
|
||||
const tools: ToolSet = params.tools
|
||||
console.log('tools', tools)
|
||||
// storeTools(context, tools)
|
||||
storeGlobalTools(executionId, tools)
|
||||
|
||||
// 构建系统提示符
|
||||
const userSystemPrompt = typeof params.system === 'string' ? params.system : ''
|
||||
const systemPrompt = await buildSystemPrompt(userSystemPrompt, tools)
|
||||
|
||||
// 将工具信息保存到参数中（用于后续解析）
|
||||
const transformedParams = {
|
||||
...params,
|
||||
system: systemPrompt,
|
||||
// 移除 tools，改为 prompt 模式
|
||||
tools: undefined
|
||||
}
|
||||
console.log('transformedParams', transformedParams)
|
||||
return transformedParams
|
||||
},
|
||||
|
||||
// 流式处理：监听 step-finish 事件并处理工具调用
|
||||
transformStream: (_, context: AiRequestContext) => () => {
|
||||
let textBuffer = ''
|
||||
let executedResults: { toolCallId: string; toolName: string; result: any; isError?: boolean }[] = []
|
||||
|
||||
return new TransformStream<any>({
|
||||
async transform(chunk, controller) {
|
||||
console.log('chunk', chunk)
|
||||
// 收集文本内容
|
||||
if (chunk.type === 'text-delta') {
|
||||
textBuffer += chunk.textDelta || ''
|
||||
console.log('textBuffer', textBuffer)
|
||||
controller.enqueue(chunk)
|
||||
return
|
||||
}
|
||||
|
||||
// 监听 step-finish 事件
|
||||
if (chunk.type === 'step-finish' || chunk.type === 'finish') {
|
||||
console.log('[MCP Prompt Stream] Received step-finish, checking for tool use...')
|
||||
|
||||
// 获取工具信息
|
||||
const tools = getGlobalTools(executionId)
|
||||
console.log('tools', tools)
|
||||
if (!tools) {
|
||||
console.log('[MCP Prompt Stream] No tools available, passing through')
|
||||
controller.enqueue(chunk)
|
||||
return
|
||||
}
|
||||
|
||||
// 解析工具调用
|
||||
const parsedTools = parseToolUse(textBuffer, tools)
|
||||
// console.log('textBuffer', textBuffer)
|
||||
const validToolUses = parsedTools.filter((t) => t.status === 'pending')
|
||||
console.log('parsedTools', parsedTools)
|
||||
|
||||
// 如果没有有效的工具调用，直接传递原始事件
|
||||
if (validToolUses.length === 0) {
|
||||
console.log('[MCP Prompt Stream] No valid tool uses found, passing through')
|
||||
controller.enqueue(chunk)
|
||||
return
|
||||
}
|
||||
|
||||
console.log('[MCP Prompt Stream] Found valid tool uses:', validToolUses.length)
|
||||
|
||||
// 修改 step-finish 事件，标记为工具调用
|
||||
if (chunk.type !== 'finish') {
|
||||
controller.enqueue({
|
||||
...chunk,
|
||||
finishReason: 'tool-call'
|
||||
})
|
||||
}
|
||||
|
||||
// 发送 step-start 事件（工具调用步骤开始）
|
||||
controller.enqueue({
|
||||
type: 'step-start'
|
||||
})
|
||||
|
||||
// 执行工具调用
|
||||
executedResults = []
|
||||
for (const toolUse of validToolUses) {
|
||||
try {
|
||||
const tool = tools[toolUse.toolName]
|
||||
if (!tool || typeof tool.execute !== 'function') {
|
||||
throw new Error(`Tool "${toolUse.toolName}" has no execute method`)
|
||||
}
|
||||
|
||||
console.log(`[MCP Prompt Stream] Executing tool: ${toolUse.toolName}`, toolUse.arguments)
|
||||
// 发送 tool-call 事件
|
||||
controller.enqueue({
|
||||
type: 'tool-call',
|
||||
toolCallId: toolUse.id,
|
||||
toolName: toolUse.toolName,
|
||||
args: toolUse.arguments
|
||||
})
|
||||
|
||||
const result = await tool.execute(toolUse.arguments, {
|
||||
toolCallId: toolUse.id,
|
||||
messages: [],
|
||||
abortSignal: new AbortController().signal
|
||||
})
|
||||
|
||||
// 发送 tool-result 事件
|
||||
controller.enqueue({
|
||||
type: 'tool-result',
|
||||
toolCallId: toolUse.id,
|
||||
toolName: toolUse.toolName,
|
||||
args: toolUse.arguments,
|
||||
result
|
||||
})
|
||||
|
||||
executedResults.push({
|
||||
toolCallId: toolUse.id,
|
||||
toolName: toolUse.toolName,
|
||||
result,
|
||||
isError: false
|
||||
})
|
||||
} catch (error) {
|
||||
console.error(`[MCP Prompt Stream] Tool execution failed: ${toolUse.toolName}`, error)
|
||||
|
||||
controller.enqueue({
|
||||
type: 'tool-result',
|
||||
toolCallId: toolUse.id,
|
||||
toolName: toolUse.toolName,
|
||||
args: toolUse.arguments,
|
||||
isError: true,
|
||||
result: error instanceof Error ? error.message : String(error)
|
||||
})
|
||||
|
||||
executedResults.push({
|
||||
toolCallId: toolUse.id,
|
||||
toolName: toolUse.toolName,
|
||||
result: error instanceof Error ? error.message : String(error),
|
||||
isError: true
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
// 发送最终的 step-finish 事件
|
||||
controller.enqueue({
|
||||
type: 'step-finish',
|
||||
finishReason: 'tool-call'
|
||||
// usage: { completionTokens: 0, promptTokens: 0, totalTokens: 0 }
|
||||
})
|
||||
|
||||
// 递归调用逻辑
|
||||
if (context.recursiveCall && validToolUses.length > 0) {
|
||||
console.log('[MCP Prompt] Starting recursive call after tool execution...')
|
||||
|
||||
// 构建工具结果的文本表示，使用Cherry Studio标准格式
|
||||
const toolResultsText = executedResults
|
||||
.map((tr) => {
|
||||
if (!tr.isError) {
|
||||
return `<tool_use_result>\n <name>${tr.toolName}</name>\n <result>${JSON.stringify(tr.result)}</result>\n</tool_use_result>`
|
||||
} else {
|
||||
const error = tr.result || 'Unknown error'
|
||||
return `<tool_use_result>\n <name>${tr.toolName}</name>\n <error>${error}</error>\n</tool_use_result>`
|
||||
}
|
||||
})
|
||||
.join('\n\n')
|
||||
|
||||
// 构建新的对话消息
|
||||
const newMessages = [
|
||||
...(context.originalParams.messages || []),
|
||||
{
|
||||
role: 'assistant',
|
||||
content: textBuffer
|
||||
},
|
||||
{
|
||||
role: 'user',
|
||||
content: toolResultsText
|
||||
}
|
||||
]
|
||||
|
||||
// 递归调用，继续对话
|
||||
const recursiveParams = {
|
||||
...context.originalParams,
|
||||
messages: newMessages,
|
||||
tools: tools // 重新传递 tools
|
||||
}
|
||||
|
||||
try {
|
||||
const recursiveResult = await context.recursiveCall(recursiveParams)
|
||||
|
||||
// 将递归调用的结果流接入当前流
|
||||
if (recursiveResult && recursiveResult.fullStream) {
|
||||
const reader = recursiveResult.fullStream.getReader()
|
||||
try {
|
||||
while (true) {
|
||||
const { done, value } = await reader.read()
|
||||
if (done) {
|
||||
break
|
||||
}
|
||||
|
||||
// 将递归流的数据传递到当前流
|
||||
controller.enqueue(value)
|
||||
}
|
||||
} finally {
|
||||
reader.releaseLock()
|
||||
}
|
||||
} else {
|
||||
console.warn('[MCP Prompt] No fullstream found in recursive result:', recursiveResult)
|
||||
}
|
||||
} catch (error) {
|
||||
console.error('[MCP Prompt] Recursive call failed:', error)
|
||||
// 发送错误信息后也要确保流不会中断
|
||||
controller.enqueue({
|
||||
type: 'text-delta',
|
||||
textDelta: `\n\n[Error: Recursive call failed: ${error instanceof Error ? error.message : String(error)}]`
|
||||
})
|
||||
|
||||
// 发送一个错误后的结束信号
|
||||
controller.enqueue({
|
||||
type: 'finish',
|
||||
finishReason: 'error'
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
// 清理状态
|
||||
// clearGlobalTools(executionId)
|
||||
textBuffer = ''
|
||||
executedResults = []
|
||||
return
|
||||
}
|
||||
|
||||
// 对于其他类型的事件，直接传递
|
||||
controller.enqueue(chunk)
|
||||
},
|
||||
|
||||
flush() {
|
||||
// 清理全局存储
|
||||
clearGlobalTools(executionId)
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
// transformResult: async (result: any, context: AiRequestContext) => {
|
||||
// // 这个方法现在主要用于非流式场景
|
||||
// if (!enabled || !result || typeof result.text !== 'string') return result
|
||||
|
||||
// console.log('[MCP Prompt] transformResult called - likely non-streaming mode')
|
||||
|
||||
// // 从 WeakMap 中获取工具信息
|
||||
// const tools: ToolSet | undefined = getStoredTools(context)
|
||||
// if (!tools || typeof tools !== 'object') return result
|
||||
|
||||
// // 使用工具解析函数（默认或自定义）
|
||||
// const parsedTools = parseToolUse(result.text, tools)
|
||||
// if (!parsedTools || parsedTools.length === 0) return result
|
||||
|
||||
// // 过滤掉解析失败的工具调用
|
||||
// const validToolUses = parsedTools.filter((t) => t.status === 'pending')
|
||||
// if (validToolUses.length === 0) {
|
||||
// console.warn('[MCP Prompt] No valid tool uses found:', parsedTools)
|
||||
// return result
|
||||
// }
|
||||
|
||||
// // 只在非流式模式下执行工具调用并递归
|
||||
// if (context.recursiveCall) {
|
||||
// console.log('[MCP Prompt] Non-streaming: Executing tools and continuing conversation...')
|
||||
|
||||
// // 执行工具调用
|
||||
// const toolResults = await Promise.all(
|
||||
// validToolUses.map(async (toolUse) => {
|
||||
// try {
|
||||
// const tool = tools[toolUse.toolName]
|
||||
// if (!tool || typeof tool.execute !== 'function') {
|
||||
// throw new Error(`Tool "${toolUse.toolName}" has no execute method`)
|
||||
// }
|
||||
|
||||
// console.log(`[MCP Prompt] Non-streaming: Executing tool: ${toolUse.toolName}`, toolUse.arguments)
|
||||
|
||||
// const result = await tool.execute(toolUse.arguments, {
|
||||
// toolCallId: toolUse.id,
|
||||
// messages: [],
|
||||
// abortSignal: new AbortController().signal
|
||||
// })
|
||||
|
||||
// return {
|
||||
// id: toolUse.id,
|
||||
// name: toolUse.toolName,
|
||||
// arguments: toolUse.arguments,
|
||||
// result,
|
||||
// success: true
|
||||
// }
|
||||
// } catch (error) {
|
||||
// console.error(`[MCP Prompt] Non-streaming: Tool execution failed: ${toolUse.toolName}`, error)
|
||||
// return {
|
||||
// id: toolUse.id,
|
||||
// name: toolUse.toolName,
|
||||
// arguments: toolUse.arguments,
|
||||
// error: error instanceof Error ? error.message : String(error),
|
||||
// success: false
|
||||
// }
|
||||
// }
|
||||
// })
|
||||
// )
|
||||
|
||||
// // 构建工具结果的文本表示
|
||||
// const toolResultsText = toolResults
|
||||
// .map((tr) => {
|
||||
// if (tr.success) {
|
||||
// return `<tool_use_result>\n <name>${tr.name}</name>\n <result>${JSON.stringify(tr.result)}</result>\n</tool_use_result>`
|
||||
// } else {
|
||||
// return `<tool_use_result>\n <name>${tr.name}</name>\n <error>${tr.error}</error>\n</tool_use_result>`
|
||||
// }
|
||||
// })
|
||||
// .join('\n\n')
|
||||
|
||||
// // 构建新的对话消息
|
||||
// const newMessages = [
|
||||
// ...(context.originalParams.messages || []),
|
||||
// {
|
||||
// role: 'assistant',
|
||||
// content: result.text
|
||||
// },
|
||||
// {
|
||||
// role: 'user',
|
||||
// content: toolResultsText
|
||||
// }
|
||||
// ]
|
||||
|
||||
// // 递归调用，继续对话
|
||||
// const recursiveParams = {
|
||||
// ...context.originalParams,
|
||||
// messages: newMessages,
|
||||
// tools: tools // 重新传递 tools，在新的 context 中会重新存储
|
||||
// }
|
||||
|
||||
// try {
|
||||
// console.log('[MCP Prompt] Non-streaming: Starting recursive call...')
|
||||
// const recursiveResult = await context.recursiveCall(recursiveParams)
|
||||
// return recursiveResult
|
||||
// } catch (error) {
|
||||
// console.error('[MCP Prompt] Non-streaming: Recursive call failed:', error)
|
||||
// return result
|
||||
// }
|
||||
// }
|
||||
|
||||
// return result
|
||||
// }
|
||||
}
|
||||
})
|
||||
@ -13,7 +13,8 @@ export function createContext(providerId: string, modelId: string, originalParam
|
||||
originalParams,
|
||||
metadata: {},
|
||||
startTime: Date.now(),
|
||||
requestId: `${providerId}-${modelId}-${Date.now()}-${Math.random().toString(36).slice(2)}`
|
||||
requestId: `${providerId}-${modelId}-${Date.now()}-${Math.random().toString(36).slice(2)}`,
|
||||
recursiveCall: undefined
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -1,5 +1,3 @@
|
||||
import type { TextStreamPart, ToolSet } from 'ai'
|
||||
|
||||
import { AiPlugin, AiRequestContext } from './types'
|
||||
|
||||
/**
|
||||
@ -121,18 +119,8 @@ export class PluginManager {
|
||||
/**
|
||||
* 收集所有流转换器（返回数组，AI SDK 原生支持）
|
||||
*/
|
||||
collectStreamTransforms<TOOLS extends ToolSet>(): Array<
|
||||
(options: {
|
||||
tools?: TOOLS
|
||||
stopStream: () => void
|
||||
}) => TransformStream<TextStreamPart<TOOLS>, TextStreamPart<TOOLS>>
|
||||
> {
|
||||
return this.plugins.map((plugin) => plugin.transformStream).filter(Boolean) as Array<
|
||||
(options: {
|
||||
tools?: TOOLS
|
||||
stopStream: () => void
|
||||
}) => TransformStream<TextStreamPart<TOOLS>, TextStreamPart<TOOLS>>
|
||||
>
|
||||
collectStreamTransforms(params: any, context: AiRequestContext) {
|
||||
return this.plugins.map((plugin) => plugin.transformStream?.(params, context))
|
||||
}
|
||||
|
||||
/**
|
||||
|
||||
@ -1,5 +1,11 @@
|
||||
import type { TextStreamPart, ToolSet } from 'ai'
|
||||
|
||||
/**
|
||||
* 递归调用函数类型
|
||||
* 使用 any 是因为递归调用时参数和返回类型可能完全不同
|
||||
*/
|
||||
export type RecursiveCallFn = (newParams: any) => Promise<any>
|
||||
|
||||
/**
|
||||
* AI 请求上下文
|
||||
*/
|
||||
@ -10,6 +16,8 @@ export interface AiRequestContext {
|
||||
metadata: Record<string, any>
|
||||
startTime: number
|
||||
requestId: string
|
||||
recursiveCall?: RecursiveCallFn
|
||||
[key: string]: any
|
||||
}
|
||||
|
||||
/**
|
||||
@ -33,7 +41,10 @@ export interface AiPlugin {
|
||||
onError?: (error: Error, context: AiRequestContext) => void | Promise<void>
|
||||
|
||||
// 【Stream】流处理 - 直接使用 AI SDK
|
||||
transformStream?: <TOOLS extends ToolSet>(options: {
|
||||
transformStream?: (
|
||||
params: any,
|
||||
context: AiRequestContext
|
||||
) => <TOOLS extends ToolSet>(options: {
|
||||
tools: TOOLS
|
||||
stopStream: () => void
|
||||
}) => TransformStream<TextStreamPart<TOOLS>, TextStreamPart<TOOLS>>
|
||||
|
||||
@ -68,6 +68,12 @@ export class PluginEngine<T extends ProviderId = ProviderId> {
|
||||
// 使用正确的createContext创建请求上下文
|
||||
const context = createContext(this.providerId, modelId, params)
|
||||
|
||||
// 🔥 为上下文添加递归调用能力
|
||||
context.recursiveCall = (newParams: any): Promise<TResult> => {
|
||||
// 递归调用自身，重新走完整的插件流程
|
||||
return this.executeWithPlugins(methodName, modelId, newParams, executor)
|
||||
}
|
||||
|
||||
try {
|
||||
// 1. 触发请求开始事件
|
||||
await this.pluginManager.executeParallel('onRequestStart', context)
|
||||
@ -109,6 +115,12 @@ export class PluginEngine<T extends ProviderId = ProviderId> {
|
||||
// 创建请求上下文
|
||||
const context = createContext(this.providerId, modelId, params)
|
||||
|
||||
// 🔥 为上下文添加递归调用能力
|
||||
context.recursiveCall = (newParams: any): Promise<TResult> => {
|
||||
// 递归调用自身，重新走完整的插件流程
|
||||
return this.executeStreamWithPlugins(methodName, modelId, newParams, executor)
|
||||
}
|
||||
|
||||
try {
|
||||
// 1. 触发请求开始事件
|
||||
await this.pluginManager.executeParallel('onRequestStart', context)
|
||||
@ -121,7 +133,7 @@ export class PluginEngine<T extends ProviderId = ProviderId> {
|
||||
const transformedParams = await this.pluginManager.executeSequential('transformParams', params, context)
|
||||
|
||||
// 4. 收集流转换器
|
||||
const streamTransforms = this.pluginManager.collectStreamTransforms()
|
||||
const streamTransforms = this.pluginManager.collectStreamTransforms(transformedParams, context)
|
||||
|
||||
// 5. 执行流式 API 调用
|
||||
const result = await executor(finalModelId, transformedParams, streamTransforms)
|
||||
|
||||
@ -15,6 +15,7 @@ import {
|
||||
type ProviderSettingsMap,
|
||||
StreamTextParams
|
||||
} from '@cherrystudio/ai-core'
|
||||
import { createMCPPromptPlugin } from '@cherrystudio/ai-core/core/plugins/built-in'
|
||||
import { isDedicatedImageGenerationModel } from '@renderer/config/models'
|
||||
import { createVertexProvider, isVertexAIConfigured, isVertexProvider } from '@renderer/hooks/useVertexAI'
|
||||
import type { GenerateImageParams, Model, Provider } from '@renderer/types'
|
||||
@ -113,7 +114,16 @@ export default class ModernAiProvider {
|
||||
// TODO:如果后续在调用completions时需要切换provider的话,
|
||||
// 初始化时不构建中间件，等到需要时再构建
|
||||
const config = providerToAiSdkConfig(provider)
|
||||
|
||||
// 创建MCP Prompt插件
|
||||
const mcpPromptPlugin = createMCPPromptPlugin({
|
||||
enabled: true
|
||||
})
|
||||
|
||||
console.log('[Modern AI Provider] Creating executor with MCP Prompt plugin enabled')
|
||||
|
||||
this.modernExecutor = createExecutor(config.providerId, config.options, [
|
||||
mcpPromptPlugin,
|
||||
reasonPlugin({
|
||||
delayInMs: 80,
|
||||
chunkingRegex: /([\u4E00-\u9FFF]{3})|\S+\s+/
|
||||
@ -174,16 +184,6 @@ export default class ModernAiProvider {
|
||||
if (middlewareConfig.onChunk) {
|
||||
// 流式处理 - 使用适配器
|
||||
const adapter = new AiSdkToChunkAdapter(middlewareConfig.onChunk)
|
||||
// this.modernExecutor.pluginEngine.use(
|
||||
// createMCPPromptPlugin({
|
||||
// mcpTools: middlewareConfig.mcpTools || [],
|
||||
// assistant: params.assistant,
|
||||
// onChunk: middlewareConfig.onChunk,
|
||||
// recursiveCall: this.modernExecutor.streamText,
|
||||
// recursionDepth: 0,
|
||||
// maxRecursionDepth: 20
|
||||
// })
|
||||
// )
|
||||
const streamResult = await this.modernExecutor.streamText(
|
||||
modelId,
|
||||
params,
|
||||
|
||||
@ -20,8 +20,8 @@ export interface MCPPromptPluginConfig {
|
||||
* 创建 MCP Prompt 模式插件
|
||||
* 支持在 prompt 模式下解析文本中的工具调用并执行
|
||||
*/
|
||||
export const createMCPPromptPlugin = (config: MCPPromptPluginConfig) => {
|
||||
return definePlugin({
|
||||
export const createMCPPromptPlugin = definePlugin((config: MCPPromptPluginConfig) => {
|
||||
return {
|
||||
name: 'mcp-prompt-plugin',
|
||||
|
||||
// 1. 参数转换 - 注入工具描述到系统提示
|
||||
@ -49,7 +49,7 @@ export const createMCPPromptPlugin = (config: MCPPromptPluginConfig) => {
|
||||
},
|
||||
|
||||
// 2. 流处理 - 检测工具调用并执行
|
||||
transformStream: () => {
|
||||
transformStream: () => () => {
|
||||
let fullResponseText = ''
|
||||
let hasProcessedTools = false
|
||||
|
||||
@ -87,8 +87,8 @@ export const createMCPPromptPlugin = (config: MCPPromptPluginConfig) => {
|
||||
}
|
||||
})
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
})
|
||||
|
||||
/**
|
||||
* 处理工具调用并执行递归
|
||||
|
||||
@ -3,7 +3,7 @@ import { definePlugin } from '@cherrystudio/ai-core'
|
||||
export default definePlugin(({ delayInMs, chunkingRegex }: { delayInMs: number; chunkingRegex: RegExp }) => ({
|
||||
name: 'reasonPlugin',
|
||||
|
||||
transformStream: () => {
|
||||
transformStream: () => () => {
|
||||
let buffer = ''
|
||||
const delay = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms))
|
||||
const detectChunk = (buffer: string) => {
|
||||
|
||||
@ -2,9 +2,10 @@ import { definePlugin, smoothStream } from '@cherrystudio/ai-core'
|
||||
|
||||
export default definePlugin({
|
||||
name: 'textPlugin',
|
||||
transformStream: smoothStream({
|
||||
delayInMs: 80,
|
||||
// 中文3个字符一个chunk,英文一个单词一个chunk
|
||||
chunking: /([\u4E00-\u9FFF]{3})|\S+\s+/
|
||||
})
|
||||
transformStream: () =>
|
||||
smoothStream({
|
||||
delayInMs: 80,
|
||||
// 中文3个字符一个chunk,英文一个单词一个chunk
|
||||
chunking: /([\u4E00-\u9FFF]{3})|\S+\s+/
|
||||
})
|
||||
})
|
||||
|
||||
@ -16,7 +16,7 @@
|
||||
"@renderer/*": ["src/renderer/src/*"],
|
||||
"@shared/*": ["packages/shared/*"],
|
||||
"@types": ["src/renderer/src/types/index.ts"],
|
||||
"@cherrystudio/ai-core": ["packages/aiCore/src/"]
|
||||
"@cherrystudio/ai-core/*": ["packages/aiCore/src/*"]
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user