#!/usr/bin/env node
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";
import { runTwoTierSearch, type EngineName } from "./engines.js";
import { fetchAndExtract } from "./extract.js";
const toInt = (v: string | undefined, def: number) => {
const n = Number(v);
return Number.isFinite(n) && n > 0 ? Math.floor(n) : def;
};
const DEFAULT_LIMIT = toInt(process.env.MAX_RESULTS, 10);
const server = new McpServer({ name: "mcp-web-search", version: "1.0.0" });
server.registerTool(
"search_web",
{
title: "Web Search (Fast: DuckDuckGo, Deep: Puppeteer/Bing)",
description:
"Two-tier web search: runs fast DuckDuckGo HTML search by default, escalates to Puppeteer/Bing if results are insufficient. No API keys required.",
inputSchema: {
q: z.string(),
limit: z.number().int().min(1).max(50).default(DEFAULT_LIMIT).optional(),
lang: z.string().default("en").optional(),
mode: z.enum(["fast", "deep", "auto"]).default("auto").optional()
}
},
async ({ q, limit = DEFAULT_LIMIT, lang = "en", mode = "auto" }) => {
const res = await runTwoTierSearch({ q, limit: Math.min(Math.max(1, limit), 50), lang, mode });
const payload = { ...res, items: res.items.slice(0, limit) };
return { content: [{ type: "text", text: JSON.stringify(payload, null, 2) }] };
}
);
server.registerTool(
"fetch_url",
{
title: "Fetch and Extract URL Content",
description:
"Fetches content from a URL (HTML/PDF) and extracts readable text. Supports truncation modes: compact (~3000 chars), standard (~8000 chars, default), full (no truncation). Output formats: markdown (default), text, html.",
inputSchema: {
url: z.string().url(),
mode: z.enum(["compact", "standard", "full"]).optional(),
max_length: z.number().int().min(1000).max(100000).optional(),
format: z.enum(["markdown", "text", "html"]).optional()
}
},
async ({ url, mode, max_length, format }) => {
const doc = await fetchAndExtract(url, { mode, max_length, format });
return { content: [{ type: "text", text: JSON.stringify(doc, null, 2) }] };
}
);
server.registerTool(
"summarize_url",
{
title: "Summarize URL Content",
description: "Fetches content from a URL and generates a concise summary.",
inputSchema: { url: z.string().url() }
},
async ({ url }) => {
const doc = await fetchAndExtract(url);
try {
const content = doc.markdown || doc.text || "";
const prompt = `Provide a concise summary (<=10 sentences) of the following content:\n\nTitle: ${doc.title || "(none)"}\nURL: ${doc.url}\n\n--- Content ---\n${content.slice(0, 12000)}`;
const resp = await (server as any).server.createMessage({
messages: [{ role: "user", content: { type: "text", text: prompt } }],
maxTokens: 800
});
const text =
resp.content && resp.content.type === "text"
? resp.content.text
: "(unable to generate summary)";
return { content: [{ type: "text", text }] };
} catch {
const fallback = (doc.markdown || doc.text || "").slice(0, 2000);
return { content: [{ type: "text", text: fallback || "(no content to summarize)" }] };
}
}
);
server.registerTool(
"wiki_get",
{
title: "Wikipedia: Get Summary",
description:
"Retrieves a Wikipedia summary for a given title. Supports multiple languages (default: en).",
inputSchema: { title: z.string(), lang: z.string().default("en").optional() }
},
async ({ title, lang = "en" }) => {
const { wikiGetSummary } = await import("./wikipedia.js");
const summary = await wikiGetSummary(title, lang);
return { content: [{ type: "text", text: JSON.stringify(summary, null, 2) }] };
}
);
server.registerTool(
"wiki_multi",
{
title: "Wikipedia: Multi-Language Summary",
description:
"Retrieves Wikipedia summaries in multiple languages for a given term. Uses langlinks to map titles accurately across languages.",
inputSchema: {
term: z.string(),
baseLang: z.string().default("en").optional(),
langs: z.array(z.string()).default(["en"]).optional()
}
},
async ({ term, baseLang = "en", langs = ["en"] }) => {
const { wikiGetMultiSummary } = await import("./wikipedia.js");
const out = await wikiGetMultiSummary(term, baseLang, langs);
return { content: [{ type: "text", text: JSON.stringify(out, null, 2) }] };
}
);
async function main() {
const transport = new StdioServerTransport();
await server.connect(transport);
console.error("mcp-web-search ready (stdio)...");
}
main().catch(err => {
console.error(err);
process.exit(1);
});
