firecrawl-mcp 3.20.3 → 3.20.4

package/dist/research.js

@@ -14,9 +14,6 @@
  */
 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)
@@ -35,6 +32,104 @@ function withQuery(path, params) {
     const qs = params.toString();
     return qs ? `${path}?${qs}` : path;
 }
+// --- result formatting (ported from research-index-front/src/agent_eval.ts) ---
+// Max authors to print per paper (with affiliations); the rest collapse to a
+// "+N more" tail so a large collaboration doesn't flood the context.
+const MAX_AUTHORS = 15;
+// Cap each abstract so a page of hits stays within the MCP output-token limit.
+const MAX_ABSTRACT_CHARS = 600;
+// Per-affiliation char cap — keeps one long org string (e.g. a full multi-dept
+// university address) from bloating the authors line.
+const MAX_AFFIL_CHARS = 60;
+// Hard ceiling on the whole authors line, as a final guard.
+const MAX_AUTHORS_LINE_CHARS = 400;
+/** Best display id for a paper: its arXiv id, falling back to the canonical id. */
+function displayId(p) {
+    return p.ids?.arxiv?.[0] ?? p.paper_id ?? '?';
+}
+/** Format the authors line, accepting either the string or structured form. */
+function fmtAuthors(authors) {
+    if (!authors)
+        return null;
+    let shown;
+    let total;
+    if (typeof authors === 'string') {
+        const names = authors
+            .split(',')
+            .map((s) => s.trim())
+            .filter(Boolean);
+        if (names.length === 0)
+            return null;
+        total = names.length;
+        shown = names.slice(0, MAX_AUTHORS);
+    }
+    else {
+        if (authors.length === 0)
+            return null;
+        total = authors.length;
+        shown = authors.slice(0, MAX_AUTHORS).map((a) => {
+            const aff = a.affiliation?.trim();
+            return aff ? `${a.name} (${aff.slice(0, MAX_AFFIL_CHARS)})` : a.name;
+        });
+    }
+    const extra = total > MAX_AUTHORS ? `; +${total - MAX_AUTHORS} more` : '';
+    return ('Authors: ' + shown.join('; ') + extra).slice(0, MAX_AUTHORS_LINE_CHARS);
+}
+/** Render ranked papers as `[id] title` / authors / abstract blocks. */
+function fmtHits(results) {
+    if (!results || results.length === 0)
+        return '(no results)';
+    return results
+        .map((r) => {
+        const lines = [`[${displayId(r)}] ${r.title ?? '(untitled)'}`];
+        const authors = fmtAuthors(r.authors);
+        if (authors)
+            lines.push(authors);
+        lines.push((r.abstract || '(no abstract)')
+            .replace(/\s+/g, ' ')
+            .slice(0, MAX_ABSTRACT_CHARS));
+        return lines.join('\n');
+    })
+        .join('\n\n');
+}
+// Cap GitHub matched content so a page of results stays within the MCP
+// output-token limit. Higher than abstracts since issue/PR threads carry the
+// signal (repro steps, stack traces) the agent actually needs to verify.
+const MAX_GITHUB_CONTENT_CHARS = 1200;
+/**
+ * Render GitHub history/readme hits as `[repo#number] (kind)` / url / body
+ * blocks — the same shape as `fmtHits`, but tuned for issues/PRs and readmes.
+ * Markdown content keeps its newlines (so tables/code survive); only readmes and
+ * snippets fall back when full content is absent.
+ */
+function fmtGithub(results) {
+    if (!results || results.length === 0)
+        return '(no results)';
+    return results
+        .map((r) => {
+        const lines = [];
+        if (r.resultType === 'repo_readme') {
+            lines.push(`[${r.repo ?? '?'}] README`);
+        }
+        else {
+            const ref = r.number != null ? `#${r.number}` : '';
+            const meta = [
+                r.pageType,
+                r.segmentCount ? `${r.segmentCount} segments` : '',
+            ]
+                .filter(Boolean)
+                .join(', ');
+            lines.push(`[${r.repo ?? '?'}${ref}]${meta ? ` (${meta})` : ''}`);
+        }
+        const url = r.readmeUrl ?? r.url;
+        if (url)
+            lines.push(url);
+        const body = (r.contentMd || r.snippet || '').trim();
+        lines.push(body ? body.slice(0, MAX_GITHUB_CONTENT_CHARS) : '(no content)');
+        return lines.join('\n');
+    })
+        .join('\n\n');
+}
 /** Only present these tools when the session has research enabled. */
 const canAccess = (session) => session?.research === true;
 export function registerResearchTools(server, getClient) {
@@ -83,7 +178,7 @@ export function registerResearchTools(server, getClient) {
             appendParam(params, 'to', to);
             const client = getClient(session);
             const res = await client.http.get(withQuery(`${BASE}/papers`, params));
-            return asText(res.data);
+            return fmtHits(res.data?.results);
         },
     });
     // --- related_papers ---
@@ -127,7 +222,8 @@ export function registerResearchTools(server, getClient) {
             appendParam(params, 'anchor', anchors);
             const client = getClient(session);
             const res = await client.http.get(withQuery(`${BASE}/papers/${encodeURIComponent(primary)}/similar`, params));
-            return asText(res.data);
+            const note = res.data?.note ? `\nnote: ${res.data.note}` : '';
+            return `${fmtHits(res.data?.results)}\n(pool_size=${res.data?.pool_size ?? 0})${note}`;
         },
     });
     // --- read_paper ---
@@ -161,11 +257,13 @@ export function registerResearchTools(server, getClient) {
             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);
+            const passages = res.data?.passages ?? [];
+            return passages.length
+                ? passages.map((p) => p.text).join('\n---\n')
+                : '(no full-text passages available for this paper)';
         },
     });
     // --- search_github ---
-    // TODO: description pending — the user is writing this one.
     server.addTool({
         name: 'firecrawl_research_search_github',
         canAccess,
@@ -187,7 +285,7 @@ export function registerResearchTools(server, getClient) {
             appendParam(params, 'k', k);
             const client = getClient(session);
             const res = await client.http.get(withQuery(`${BASE}/github`, params));
-            return asText(res.data);
+            return fmtGithub(res.data?.results);
         },
     });
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "firecrawl-mcp",
-  "version": "3.20.3",
+  "version": "3.20.4",
   "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",