@metalift/mcp - npm Package Compare versions

Comparing version 1.0.6 to 1.0.7

dist/web-search.d.ts

+export declare const WEB_SEARCH_RESULT_LIMIT = 10;
+export declare const WEB_SEARCH_CREDITS = 2;
+export declare function buildWebSearchRequest(args: {
+    query: string;
+    categories?: string[];
+    language?: string;
+}): {
+    limit: number;
+    query: string;
+    categories?: string[];
+    language?: string;
+};

dist/web-search.js

+export const WEB_SEARCH_RESULT_LIMIT = 10;
+export const WEB_SEARCH_CREDITS = 2;
+export function buildWebSearchRequest(args) {
+    return { ...args, limit: WEB_SEARCH_RESULT_LIMIT };
+}

dist/web-search.test.d.ts

+export {};

dist/web-search.test.js

+import assert from "node:assert/strict";
+import test from "node:test";
+import { buildWebSearchRequest, WEB_SEARCH_RESULT_LIMIT } from "./web-search.js";
+test("buildWebSearchRequest always uses top-10 limit", () => {
+    assert.equal(WEB_SEARCH_RESULT_LIMIT, 10);
+    assert.deepEqual(buildWebSearchRequest({ query: "docker docs" }), {
+        query: "docker docs",
+        limit: 10,
+    });
+    assert.deepEqual(buildWebSearchRequest({ query: "news", categories: ["news"], language: "en" }), { query: "news", categories: ["news"], language: "en", limit: 10 });
+});

dist/client.d.ts

@@ -22,2 +22,3 @@ import type { ScrapeArgs } from "./scrape-args.js";
     map(params: Record<string, unknown>): Promise<Record<string, unknown> & BillingMeta>;
+    search(params: Record<string, unknown>): Promise<Record<string, unknown> & BillingMeta>;
     jobStatus(jobId: string): Promise<Record<string, unknown> & BillingMeta>;
@@ -24,0 +25,0 @@ listStrategies(): Promise<{

dist/client.js

@@ -99,2 +99,5 @@ const B2B_ATTESTATION_HEADER = "X-B2B-Attestation";
     }
+    search(params) {
+        return this.request("/v1/search", { method: "POST", body: JSON.stringify(params) });
+    }
     jobStatus(jobId) {
@@ -101,0 +104,0 @@ return this.request(`/v1/jobs/${jobId}`);

dist/index.js

@@ -6,12 +6,28 @@ #!/usr/bin/env node
 import { MetaliftClient } from "./client.js";
+import { buildWebSearchRequest, WEB_SEARCH_RESULT_LIMIT } from "./web-search.js";
 import { normalizeScrapeArgs } from "./scrape-args.js";
 const COMPLIANCE_NOTICE = "You are solely responsible for complying with website terms, robots.txt, copyright, and data protection laws when using scraped content.";
+const SERVER_INSTRUCTIONS = `Metalift provides web search and web scraping as separate, independently billed tools.
+
+Web search (metalift_web_search):
+- Returns SERP metadata only: title, url, snippet, engine, score — not full page content.
+- Costs 2 credits per successful search (flat fee).
+- Always returns up to ${WEB_SEARCH_RESULT_LIMIT} results.
+- Do NOT auto-scrape all search results. Review snippets first, then call metalift_scrape only for URLs that need full content.
+
+Scraping (metalift_scrape, metalift_batch_scrape):
+- Fetches page content (markdown, html, text). Billed per URL (static=1, JS=5, premium=10+ credits).
+- Use after search when the user or agent needs full page content from specific URLs.
+
+Recommended agent workflow: metalift_web_search → pick 0–N relevant URLs from snippets → metalift_scrape chosen URLs only.`;
 const client = new MetaliftClient();
 const server = new McpServer({
     name: "metalift",
-    version: "1.0.0",
+    version: "1.0.7",
+}, {
+    instructions: SERVER_INSTRUCTIONS,
 });
 server.registerTool("metalift_scrape", {
     title: "Scrape URL",
-    description: `Scrape a single URL into markdown, HTML, or text for LLM context. Default: fast direct static article extraction (strategy=article, render=static, proxy=direct, 10s timeout). For full page HTML on static sites use strategy=download with formats=["html"] (1 credit, all tiers). strategy=raw and full-page HTML without download require Enterprise tier. For WAF, SPA, retail, or JS-heavy pages pass strategy=auto or a specific strategy (spa, cloudflare, retail). Response includes credits_charged based on actual usage (static=1, JS=5, premium=10+). ${COMPLIANCE_NOTICE}`,
+    description: `Scrape a single URL into markdown, HTML, or text for LLM context. Separate from web search — use after metalift_web_search when full page content is needed. Default: fast direct static article extraction (strategy=article, render=static, proxy=direct, 10s timeout). For full page HTML on static sites use strategy=download with formats=["html"] (1 credit, all tiers). strategy=raw and full-page HTML without download require Enterprise tier. For WAF, SPA, retail, or JS-heavy pages pass strategy=auto or a specific strategy (spa, cloudflare, retail). Response includes credits_charged based on actual usage (static=1, JS=5, premium=10+). ${COMPLIANCE_NOTICE}`,
     inputSchema: {
@@ -97,2 +113,17 @@ url: z.string().url(),
 });
+server.registerTool("metalift_web_search", {
+    title: "Web Search",
+    description: `Search the web and return up to ${WEB_SEARCH_RESULT_LIMIT} SERP results (title, url, snippet, engine, score). Costs 2 credits per search. Returns search snippets only — not page content. Do not auto-scrape results; pick relevant URLs and call metalift_scrape separately for full content.`,
+    inputSchema: {
+        query: z.string().min(1).max(512),
+        categories: z.array(z.string()).optional(),
+        language: z.string().max(16).optional(),
+    },
+    annotations: { readOnlyHint: true },
+}, async (args) => {
+    const result = await client.search(buildWebSearchRequest(args));
+    return {
+        content: [{ type: "text", text: JSON.stringify(result, null, 2) }],
+    };
+});
 server.registerTool("metalift_job_status", {
@@ -168,2 +199,3 @@ title: "Get Job Status",
                         "metalift_map",
+                        "metalift_web_search",
                         "metalift_job_status",
@@ -174,2 +206,7 @@ "metalift_list_strategies",
                     ],
+                    web_search: {
+                        credits_per_search: 2,
+                        result_limit: WEB_SEARCH_RESULT_LIMIT,
+                        decoupled_from_scrape: true,
+                    },
                 }, null, 2),
@@ -248,2 +285,23 @@ },
 }));
+server.registerPrompt("research_topic", {
+    title: "Research Topic",
+    description: "Search the web for a topic, then scrape only the most relevant URLs (decoupled search + scrape workflow)",
+    argsSchema: {
+        query: z.string().describe("Search query"),
+    },
+}, async ({ query }) => ({
+    messages: [
+        {
+            role: "user",
+            content: {
+                type: "text",
+                text: `Research "${query}" using this decoupled workflow:
+1. Call metalift_web_search with query "${query}" (2 credits, top 10 SERP snippets).
+2. Review titles and snippets — do NOT scrape every result.
+3. Call metalift_scrape only for the 1–3 URLs that need full page content.
+4. Synthesize an answer citing sources.`,
+            },
+        },
+    ],
+}));
 async function main() {
@@ -250,0 +308,0 @@ const transport = new StdioServerTransport();

package.json

 {
   "name": "@metalift/mcp",
   "mcpName": "io.github.MetaLift-AI/metalift",
-  "version": "1.0.6",
+  "version": "1.0.7",
   "description": "Metalift MCP server for AI agents",
@@ -16,3 +16,3 @@ "license": "SEE LICENSE IN LICENSE",
     "dev": "tsx src/index.ts",
-    "test": "npm run build && node --test dist/client.test.js dist/scrape-args.test.js",
+    "test": "npm run build && node --test dist/client.test.js dist/scrape-args.test.js dist/web-search.test.js",
     "prepublishOnly": "npm run build && node ../../scripts/npm-prepare-publish.mjs strip",
@@ -19,0 +19,0 @@ "postpublish": "node ../../scripts/npm-prepare-publish.mjs restore"

README.md

@@ -47,2 +47,11 @@ # @metalift/mcp
 
+## Web search vs scrape
+
+Search and scrape are **separate tools** with separate billing:
+
+1. **`metalift_web_search`** — returns up to 10 search snippets (2 credits). Review titles and snippets first.
+2. **`metalift_scrape`** — fetches full page content for URLs you choose (1+ credits per URL).
+
+Do not auto-scrape every search result. See [Web search](../../packages/platform-web/docs/web-search.md) for agent workflow examples.
+
 ## Environment variables
@@ -63,2 +72,3 @@
 | `metalift_map` | Discover site URLs |
+| `metalift_web_search` | Web search — top 10 SERP results (title, url, snippet). **2 credits** per search. Does not fetch page content; use `metalift_scrape` separately for URLs you need |
 | `metalift_job_status` | Poll async jobs |
@@ -65,0 +75,0 @@ | `metalift_list_strategies` | List scrape strategies |

No alert changes

Improved metrics

Total package byte prevSize

35563

15.77%

Number of package files

17

30.77%

Lines of code

706

14.61%

Number of lines in readme file

77

14.93%

No dependency changes