import { z } from "zod";
import { createErrorResponse, createJsonResponse, createSuccessResponse } from "../utils/mcpResponse";
import { McpToolsProvider } from "./baseToolProvider";
import { debugPush, errorPush, logPush } from "@/logger";
import { useConsumer, useProvider } from "@/utils/indexerHelper";
import { lang } from "@/utils/lang";
export class DocVectorSearchProvider extends McpToolsProvider<any> {
async getTools(): Promise<McpTool<any>[]> {
const indexProvider = useProvider();
const healthResult = await indexProvider.health();
if (healthResult == null) {
logPush("Connection with RAG backend ERROR: (RAG Tool will not be load to MCP server)", healthResult);
return [];
}
const EXPORT_API = async (question)=>{
const provider = useProvider();
return await provider.query(question)
};
if (window["OpaqueGlassAPI"]) {
window["OpaqueGlassAPI"]["ragQuery"] = EXPORT_API;
} else {
window["OpaqueGlassAPI"] = {
"ragQuery": EXPORT_API
}
}
window["OpaqueGlassAPI"][""]
return [{
name: "siyuan_generate_answer_with_doc",
description: 'This tool provides a Retrieval-Augmented Generation (RAG) based Q&A capability. It generates context-aware answers using only the notes that the user has explicitly indexed from their siyuan-notes. Please note: the tool does not access or use all documents—only those that have been indexed by the user. ',
schema: {
question: z.string().describe("Describe question about note here"),
},
handler: answerWithRAG,
title: lang("tool_title_generate_answer_with_doc"),
annotations: {
readOnlyHint: false,
destructiveHint: false,
idempotentHint: false,
}
}];
}
}
async function answerWithRAG(params, extra) {
const { question } = params;
debugPush("API被调用：RAG");
const provider = useProvider();
let progressInterval;
let timeoutId;
let finished = false;
const progressToken = extra?._meta?.progressToken;
let currentProgress = 0;
const maxDuration = 120 * 1000; // 120秒
const updateInterval = 3000; // 3秒
const progressIncrement = updateInterval / maxDuration;
if (progressToken) {
progressInterval = setInterval(() => {
currentProgress += progressIncrement;
if (currentProgress < 0.95) {
extra.sendNotification && extra.sendNotification({
method: "notifications/progress",
params: { progress: currentProgress, progressToken }
});
}
}, updateInterval);
timeoutId = setTimeout(() => {
if (!finished) {
finished = true;
clearInterval(progressInterval);
extra.sendNotification && extra.sendNotification({
method: "notifications/progress",
params: { progress: 1, progressToken }
});
}
}, maxDuration);
}
try {
const resultPromise = provider.query(question);
const result = await Promise.race([
resultPromise,
new Promise((_, reject) => setTimeout(() => reject(new Error("RAG query timeout (120s)")), maxDuration))
]);
finished = true;
if (progressInterval) clearInterval(progressInterval);
if (timeoutId) clearTimeout(timeoutId);
if (progressToken) {
extra.sendNotification && extra.sendNotification({
method: "notifications/progress",
params: { progress: 1, progressToken }
});
}
logPush("RAG result", result);
return createJsonResponse(result);
} catch (err) {
finished = true;
if (progressInterval) clearInterval(progressInterval);
if (timeoutId) clearTimeout(timeoutId);
if (progressToken) {
extra.sendNotification && extra.sendNotification({
method: "notifications/progress",
params: { progress: 1, progressToken }
});
}
errorPush("RAG API error", err);
return createErrorResponse("The tool call failed. " + (err?.message || "There was a problem with the connection to the RAG service. Please remind the user to troubleshoot the problem."));
}
}