Resolve .md page URLs in the MCP getPage tool (RND-13634) (#4735)

.changeset/mcp-getpage-markdown-urls.md

@@ -0,0 +1,5 @@
+---
+"gitbook": patch
+---
+
+Fix the MCP `getPage` tool returning "Page not found" for the Markdown URL of a page (`<page>.md`).

packages/gitbook/src/lib/pages.test.ts

@@ -50,6 +50,16 @@ describe('extractPagePath', () => {
     it('returns empty string when URL equals root base URL', () => {
         expect(extractPagePath('https://docs.example.com/', 'https://docs.example.com/')).toBe('');
     });
+
+    it('extracts path from the markdown URL of a page', () => {
+        expect(
+            extractPagePath('https://docs.example.com/api/guides/installation.md', baseURL)
+        ).toBe('guides/installation');
+        expect(extractPagePath('https://docs.example.com/api/.md', baseURL)).toBe('');
+        expect(extractPagePath('https://docs.example.com/api/notes.md.html', baseURL)).toBe(
+            'notes.md.html'
+        );
+    });
 });
 
 describe('resolveFirstDocument', () => {

packages/gitbook/src/lib/pages.ts

@@ -412,7 +412,8 @@ function flattenPages(
  * Returns the path segment after the base, or undefined if the URL doesn't match.
  */
 export function extractPagePath(url: string, baseURL: string): string | undefined {
-    const urlPath = getURLPathname(url);
+    // Pages are also served as Markdown at `<page>.md`, the form llms.txt links to.
+    const urlPath = getURLPathname(url)?.replace(/\.md$/, '');
     const basePath = getURLPathname(baseURL);
 
     if (urlPath === undefined || basePath === undefined) {

packages/gitbook/tests/mcp.test.ts

@@ -55,6 +55,29 @@ it(
     { timeout: 15_000 }
 );
 
+it(
+    'should get a page through MCP from its markdown URL',
+    async () => {
+        const client = await connectMCPClient(
+            getContentTestURL(
+                'https://gitbook-open-e2e-sites.gitbook.io/api-multi-versions-share-links/8tNo6MeXg7CkFMzSSz81/~gitbook/mcp/auth'
+            )
+        );
+
+        const response = await client.callTool({
+            name: 'getPage',
+            arguments: {
+                url: 'https://gitbook-open-e2e-sites.gitbook.io/api-multi-versions-share-links/8tNo6MeXg7CkFMzSSz81/3.0/other-page.md',
+            },
+        });
+
+        expect(response.isError).toBeFalsy();
+        // @ts-expect-error - response.content is of type unknown
+        expect(response.content[0]?.text).toContain('# Other Page');
+    },
+    { timeout: 15_000 }
+);
+
 describe('MCP on a site behind visitor authentication', () => {
     const VA_SITE_URL = 'https://gitbook-open-e2e-sites.gitbook.io/va-site-redirects-fallback';