firecrawl-mcp 3.20.1 → 3.20.3

package/README.md

@@ -428,6 +428,7 @@ Scrape content from a single URL with advanced options.
 ```
 
 **Branding format:** Extracts comprehensive brand identity (colors, fonts, typography, spacing, logo, UI components) for design analysis or style replication.
+**Privacy:** Set `redactPII: true` to return content with personally identifiable information redacted.
 
 **Returns:**
 
@@ -565,7 +566,8 @@ Search the web and optionally extract content from search results.
     "country": "us",
     "scrapeOptions": {
       "formats": ["markdown"],
-      "onlyMainContent": true
+      "onlyMainContent": true,
+      "redactPII": true
     }
   }
 }

package/dist/index.js

@@ -6,7 +6,30 @@ import { readFile } from 'node:fs/promises';
 import path from 'node:path';
 import { z } from 'zod';
 import { registerMonitorTools } from './monitor.js';
+import { registerResearchTools } from './research.js';
 dotenv.config({ debug: false, quiet: true });
+/**
+ * Decide whether the research tools should be visible for a session.
+ * Local/stdio/self-hosted: gated by `FIRECRAWL_RESEARCH=true`.
+ * Remote (HTTP): additionally enabled by a `?research=true` query param on the
+ * incoming MCP request URL.
+ */
+function isResearchEnabled(request) {
+    if (process.env.FIRECRAWL_RESEARCH === 'true')
+        return true;
+    const url = request?.url;
+    if (url) {
+        try {
+            const research = new URL(url, 'http://localhost').searchParams.get('research');
+            if (research === 'true')
+                return true;
+        }
+        catch {
+            // malformed URL — fall through to disabled
+        }
+    }
+    return false;
+}
 function normalizeHeader(value) {
     if (value == null)
         return undefined;
@@ -187,6 +210,7 @@ const server = new FastMCP({
         protectedResourceMetadataUrl: getOAuthProtectedResourceMetadataUrl(),
     },
     authenticate: async (request) => {
+        const research = isResearchEnabled(request);
         // FastMCP invokes `authenticate(undefined)` for the stdio transport
         // because there is no HTTP request context. Without this null guard,
         // accessing `request.headers` throws a TypeError, FastMCP silently
@@ -199,9 +223,9 @@ const server = new FastMCP({
         const envCred = resolveCredentialFromEnv();
         if (process.env.CLOUD_SERVICE === 'true') {
             if (!headerCred) {
-                throw new Error('Firecrawl credentials required: OAuth access token (Authorization: Bearer fco_…) or API key (x-firecrawl-api-key)');
+                throw new Error('Firecrawl credentials required: OAuth access token (Authorization: Bearer fco_...) or API key (x-firecrawl-api-key)');
             }
-            return { firecrawlApiKey: headerCred };
+            return { firecrawlApiKey: headerCred, research };
         }
         const credential = headerCred ?? envCred;
         // Self-hosted / stdio / HTTP streamable — headers supply MCP OAuth token when present
@@ -213,10 +237,10 @@ const server = new FastMCP({
             process.exit(1);
         }
         if (httpStreaming && !credential && !process.env.FIRECRAWL_API_URL) {
-            console.error('HTTP MCP transport requires FIRECRAWL_API_URL and/or credentials (OAuth: Authorization Bearer fco_…, or FIRECRAWL_API_KEY / FIRECRAWL_OAUTH_TOKEN)');
+            console.error('HTTP MCP transport requires FIRECRAWL_API_URL and/or credentials (OAuth: Authorization Bearer fco_..., or FIRECRAWL_API_KEY / FIRECRAWL_OAUTH_TOKEN)');
             process.exit(1);
         }
-        return { firecrawlApiKey: credential };
+        return { firecrawlApiKey: credential, research };
     },
     // Lightweight health endpoint for LB checks
     health: {
@@ -380,6 +404,7 @@ const scrapeParamsSchema = z.object({
     })
         .optional(),
     onlyMainContent: z.boolean().optional(),
+    redactPII: z.boolean().optional(),
     includeTags: z.array(z.string()).optional(),
     excludeTags: z.array(z.string()).optional(),
     waitFor: z.number().optional(),
@@ -525,6 +550,7 @@ If JSON extraction returns empty, minimal, or just navigation content, the page
 **Branding format:** Extracts comprehensive brand identity (colors, fonts, typography, spacing, logo, UI components) for design analysis or style replication.
 **Performance:** Add maxAge parameter for 500% faster scrapes using cached data.
 **Lockdown mode:** Set \`lockdown: true\` to serve the request only from the existing index/cache without any outbound network request. For air-gapped or compliance-constrained use where the request URL itself is considered sensitive. Errors on cache miss. Billed at 5 credits.
+**Privacy:** Set \`redactPII: true\` to return content with personally identifiable information redacted.
 **Returns:** JSON structured data, markdown, branding profile, or other formats as specified.
 ${SAFE_MODE
         ? '**Safe Mode:** Read-only content extraction. Interactive actions (click, write, executeJavascript) are disabled for security.'
@@ -1361,6 +1387,7 @@ if (process.env.CLOUD_SERVICE !== 'true') {
         })
             .optional(),
         onlyMainContent: z.boolean().optional(),
+        redactPII: z.boolean().optional(),
         includeTags: z.array(z.string()).optional(),
         excludeTags: z.array(z.string()).optional(),
         removeBase64Images: z.boolean().optional(),
@@ -1402,6 +1429,7 @@ This is the fastest and most reliable way to extract content from a document on
 
 **Supported file types:** .html, .htm, .xhtml, .pdf, .docx, .doc, .odt, .rtf, .xlsx, .xls
 **Unsupported options:** actions, screenshot/branding/changeTracking formats, waitFor > 0, location, mobile, proxy values other than "auto" or "basic".
+**Privacy:** Set \`redactPII: true\` to return content with personally identifiable information redacted.
 
 **CRITICAL - Format Selection (same rules as firecrawl_scrape):**
 When the user asks for SPECIFIC data points from a document, you MUST use JSON format with a schema. Only use markdown when the user needs the ENTIRE document content.
@@ -1538,4 +1566,18 @@ else {
     };
 }
 registerMonitorTools(server);
+// Research tools gating. FastMCP's `canAccess` is only honored on the HTTP
+// transport (the stdio path exposes every registered tool regardless), so we
+// split the two cases:
+//   - HTTP (cloud / SSE_LOCAL / HTTP_STREAMABLE_SERVER): always register; each
+//     tool's `canAccess` hides it unless the session has research enabled
+//     (`FIRECRAWL_RESEARCH=true` env or `?research=true` on the request).
+//   - stdio (local): register only when `FIRECRAWL_RESEARCH=true`, since
+//     `canAccess` cannot hide them there.
+const isHttpTransport = process.env.CLOUD_SERVICE === 'true' ||
+    process.env.SSE_LOCAL === 'true' ||
+    process.env.HTTP_STREAMABLE_SERVER === 'true';
+if (isHttpTransport || process.env.FIRECRAWL_RESEARCH === 'true') {
+    registerResearchTools(server, getClient);
+}
 await server.start(args);

package/dist/research.js

@@ -0,0 +1,193 @@
+/**
+ * Firecrawl Research tools (experimental).
+ *
+ * Thin MCP wrappers over the `/v2/research/*` endpoints (arXiv papers + GitHub
+ * history/readmes). These tools are hidden unless research is enabled for the
+ * session — locally via `FIRECRAWL_RESEARCH=true`, or remotely via the
+ * `?research=true` query param on the MCP endpoint (see `isResearchEnabled` in
+ * index.ts, which sets `session.research`).
+ *
+ * The installed `@mendable/firecrawl-js` predates the SDK's `research` client,
+ * so we call the endpoints directly through the SDK's HTTP layer (auth +
+ * retries) via `client.http.get(...)`, mirroring how the search tool reaches
+ * `/v2/search`.
+ */
+import { z } from 'zod';
+const BASE = '/v2/research';
+function asText(data) {
+    return JSON.stringify(data, null, 2);
+}
+/** Append a value (or repeated array values) to a URLSearchParams instance. */
+function appendParam(params, key, value) {
+    if (value == null)
+        return;
+    if (Array.isArray(value)) {
+        for (const v of value) {
+            if (v != null && String(v).length > 0)
+                params.append(key, String(v));
+        }
+    }
+    else {
+        params.append(key, String(value));
+    }
+}
+function withQuery(path, params) {
+    const qs = params.toString();
+    return qs ? `${path}?${qs}` : path;
+}
+/** Only present these tools when the session has research enabled. */
+const canAccess = (session) => session?.research === true;
+export function registerResearchTools(server, getClient) {
+    // --- search_papers ---
+    server.addTool({
+        name: 'firecrawl_research_search_papers',
+        canAccess,
+        annotations: {
+            title: 'Search arXiv papers',
+            readOnlyHint: true,
+            openWorldHint: true,
+        },
+        description: 'Primary entry point for finding arXiv papers by topic. Semantic (HyDE) search over arXiv ' +
+            'abstracts; returns ranked papers with arXiv id, title, and abstract. The query should be a ' +
+            'natural-language description of what you want. Run SEVERAL distinct framings of the question ' +
+            '(sibling domains, rival methods, dataset/benchmark names) rather than one query — recall ' +
+            'improves markedly with diverse framings. Returns up to `k` results (default 40).',
+        parameters: z.object({
+            query: z.string().min(1),
+            k: z.number().int().min(1).max(500).optional(),
+            authors: z
+                .array(z.string())
+                .optional()
+                .describe('Author substring filter(s); ALL must match (case-insensitive).'),
+            categories: z
+                .array(z.string())
+                .optional()
+                .describe('arXiv category filter(s) (e.g. `cs.LG`); ALL must match.'),
+            from: z
+                .string()
+                .optional()
+                .describe('Inclusive lower bound on created/updated date (`YYYY-MM-DD`).'),
+            to: z
+                .string()
+                .optional()
+                .describe('Inclusive upper bound on created/updated date (`YYYY-MM-DD`).'),
+        }),
+        execute: async (args, { session }) => {
+            const { query, k, authors, categories, from, to } = args;
+            const params = new URLSearchParams();
+            appendParam(params, 'query', query);
+            appendParam(params, 'k', k);
+            appendParam(params, 'authors', authors);
+            appendParam(params, 'categories', categories);
+            appendParam(params, 'from', from);
+            appendParam(params, 'to', to);
+            const client = getClient(session);
+            const res = await client.http.get(withQuery(`${BASE}/papers`, params));
+            return asText(res.data);
+        },
+    });
+    // --- related_papers ---
+    server.addTool({
+        name: 'firecrawl_research_related_papers',
+        canAccess,
+        annotations: {
+            title: 'Find related arXiv papers',
+            readOnlyHint: true,
+            openWorldHint: true,
+        },
+        description: 'Expand from anchor papers you have already found, via the citation graph, ranked and filtered ' +
+            'to a natural-language `intent`. Pass arXiv ids of your strongest hits as `seed_ids`. Modes: ' +
+            '`similar` (cocitation/coupling — papers in the same niche; the default), `citers` (papers ' +
+            'that cite the anchors), `references` (papers the anchors cite). This reaches relevant papers ' +
+            'that plain search misses, so use it on your best hits before finishing. A `similar` call ' +
+            'already runs a DEEP multi-round expansion internally (re-seeding from each round’s best ' +
+            'finds), so one call reaches the wider neighborhood — no need to chain many. Returns the ' +
+            'candidates plus the pool size.',
+        parameters: z.object({
+            seed_ids: z.array(z.string()).min(1).max(10),
+            intent: z.string().min(1),
+            mode: z.enum(['similar', 'citers', 'references']).optional(),
+            k: z.number().int().min(1).max(500).optional(),
+            rerank: z
+                .boolean()
+                .optional()
+                .describe('Apply an additional rerank over the fused candidates.'),
+        }),
+        execute: async (args, { session }) => {
+            const { seed_ids, intent, mode, k, rerank } = args;
+            // The endpoint takes a single primary seed in the path; any additional
+            // seeds ride along as repeated `anchor` params.
+            const [primary, ...anchors] = seed_ids;
+            const params = new URLSearchParams();
+            appendParam(params, 'intent', intent);
+            appendParam(params, 'mode', mode);
+            appendParam(params, 'k', k);
+            if (rerank != null)
+                appendParam(params, 'rerank', rerank);
+            appendParam(params, 'anchor', anchors);
+            const client = getClient(session);
+            const res = await client.http.get(withQuery(`${BASE}/papers/${encodeURIComponent(primary)}/similar`, params));
+            return asText(res.data);
+        },
+    });
+    // --- read_paper ---
+    server.addTool({
+        name: 'firecrawl_research_read_paper',
+        canAccess,
+        annotations: {
+            title: 'Read an arXiv paper',
+            readOnlyHint: true,
+            openWorldHint: true,
+        },
+        description: 'Read the most relevant in-body (full-text) passages of ONE specific paper for a question. Use ' +
+            'this to VERIFY whether a candidate actually satisfies a constraint before you include or ' +
+            "reject it (e.g. 'does this paper actually use technique X / report a score on benchmark Y'). " +
+            "Returns the best-matching passages, or a notice if the paper's full text is unavailable.",
+        parameters: z.object({
+            arxiv_id: z.string().min(1),
+            question: z.string().min(1),
+            k: z
+                .number()
+                .int()
+                .min(1)
+                .max(50)
+                .optional()
+                .describe('Number of passages to return (default 4).'),
+        }),
+        execute: async (args, { session }) => {
+            const { arxiv_id, question, k } = args;
+            const params = new URLSearchParams();
+            appendParam(params, 'query', question);
+            appendParam(params, 'k', k);
+            const client = getClient(session);
+            const res = await client.http.get(withQuery(`${BASE}/papers/${encodeURIComponent(arxiv_id)}`, params));
+            return asText(res.data);
+        },
+    });
+    // --- search_github ---
+    // TODO: description pending — the user is writing this one.
+    server.addTool({
+        name: 'firecrawl_research_search_github',
+        canAccess,
+        annotations: {
+            title: 'Search GitHub history',
+            readOnlyHint: true,
+            openWorldHint: true,
+        },
+        description: 'Search GitHub issue/PR history and repository readmes. Returns ranked matches with repo, ' +
+            'url, a short snippet, and (when available) the full matched content in markdown.',
+        parameters: z.object({
+            query: z.string().min(1),
+            k: z.number().int().min(1).max(100).optional(),
+        }),
+        execute: async (args, { session }) => {
+            const { query, k } = args;
+            const params = new URLSearchParams();
+            appendParam(params, 'query', query);
+            appendParam(params, 'k', k);
+            const client = getClient(session);
+            const res = await client.http.get(withQuery(`${BASE}/github`, params));
+            return asText(res.data);
+        },
+    });
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "firecrawl-mcp",
-  "version": "3.20.1",
+  "version": "3.20.3",
   "description": "MCP server for Firecrawl — search, scrape, and interact with the web. Supports both cloud and self-hosted instances. Features include web search, scraping, page interaction, batch processing, and LLM-powered content analysis.",
   "type": "module",
   "mcpName": "io.github.firecrawl/firecrawl-mcp-server",
@@ -15,7 +15,7 @@
   },
   "license": "MIT",
   "dependencies": {
-    "@mendable/firecrawl-js": "4.24.0",
+    "@mendable/firecrawl-js": "4.25.2",
     "dotenv": "^17.2.2",
     "firecrawl-fastmcp": "^1.0.5",
     "typescript": "^5.9.2",