Add PDF text extraction to MCP read_file and search_files

MCP's read_file previously hard-required UTF-8 text decoding, so any
PDF in an allowed folder failed with "Cannot read file as UTF-8
text" — chat attachments already supported PDFs (raw bytes to
vision-capable models), but the AI couldn't read one on its own
during agentic file-tool use. search_files' content_search had the
same gap, silently skipping PDFs.

Adds MCPService.extractPDFText(atPath:) using PDFKit (built into
macOS, no new dependency) to pull text from a PDF's text layer.
Wired into both read_file and search_files' content search. Returns
a clear error for scanned/image-only PDFs with no text layer.
Automatically covers the Research Agents sub-agent tool loop too,
since it shares the same executeTool dispatcher.

Verified live: asked the AI to read a real PDF containing "The
secret code is PINEAPPLE-42." via read_file — it extracted the text
correctly through the MCP tool-call path.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

oAI/Services/MCPService.swift

@@ -23,6 +23,7 @@
 
 import Foundation
 import os
+import PDFKit
 
 @Observable
 class MCPService {
@@ -155,7 +156,7 @@ class MCPService {
         var tools: [Tool] = [
             makeTool(
                 name: "read_file",
-                description: "Read the contents of a file. Returns the text content of the file. Maximum file size is 10MB.",
+                description: "Read the contents of a file. Returns the text content of the file. PDFs are supported (text layer is extracted automatically) — scanned/image-only PDFs with no text layer will return an error. Maximum file size is 10MB.",
                 properties: [
                     "file_path": prop("string", "The absolute path to the file to read")
                 ],
@@ -172,7 +173,7 @@ class MCPService {
             ),
             makeTool(
                 name: "search_files",
-                description: "Search for files by name pattern or content. Use 'pattern' for filename glob matching (e.g. '*.swift'). Use 'content_search' for searching inside file contents.",
+                description: "Search for files by name pattern or content. Use 'pattern' for filename glob matching (e.g. '*.swift'). Use 'content_search' for searching inside file contents (PDF text layers are searched too).",
                 properties: [
                     "pattern": prop("string", "Glob pattern to match filenames (e.g. '*.py', 'README*')"),
                     "search_path": prop("string", "Directory to search in (defaults to first allowed folder)"),
@@ -536,8 +537,17 @@ class MCPService {
             return ["error": "File too large (\(sizeMB) MB, max 10 MB)"]
         }
 
-        guard let content = try? String(contentsOfFile: resolved, encoding: .utf8) else {
+        let content: String
-            return ["error": "Cannot read file as UTF-8 text: \(filePath)"]
+        if (resolved as NSString).pathExtension.lowercased() == "pdf" {
+            guard let extracted = extractPDFText(atPath: resolved) else {
+                return ["error": "Could not extract text from PDF (it may be scanned/image-only with no text layer, encrypted, or corrupted): \(filePath)"]
+            }
+            content = extracted
+        } else {
+            guard let text = try? String(contentsOfFile: resolved, encoding: .utf8) else {
+                return ["error": "Cannot read file as UTF-8 text: \(filePath)"]
+            }
+            content = text
         }
 
         var finalContent = content
@@ -554,6 +564,19 @@ class MCPService {
         return ["content": finalContent, "path": resolved, "size": fileSize]
     }
 
+    /// Extracts plain text from a PDF's text layer, page by page. Returns nil for
+    /// scanned/image-only PDFs (no text layer), encrypted, or otherwise unreadable files.
+    func extractPDFText(atPath path: String) -> String? {
+        guard let document = PDFDocument(url: URL(fileURLWithPath: path)) else { return nil }
+        var text = ""
+        for i in 0..<document.pageCount {
+            guard let page = document.page(at: i), let pageText = page.string else { continue }
+            text += pageText + "\n\n"
+        }
+        let trimmed = text.trimmingCharacters(in: .whitespacesAndNewlines)
+        return trimmed.isEmpty ? nil : text
+    }
+
     private func listDirectory(dirPath: String, recursive: Bool) -> [String: Any] {
         let resolved = ((dirPath as NSString).expandingTildeInPath as NSString).standardizingPath
 
@@ -660,10 +683,13 @@ class MCPService {
 
             // Content search if requested
             if let searchText = contentSearch, !searchText.isEmpty {
-                guard let content = try? String(contentsOfFile: fullPath, encoding: .utf8) else {
+                let content: String?
-                    continue
+                if (fullPath as NSString).pathExtension.lowercased() == "pdf" {
+                    content = extractPDFText(atPath: fullPath)
+                } else {
+                    content = try? String(contentsOfFile: fullPath, encoding: .utf8)
                 }
-                if !content.localizedCaseInsensitiveContains(searchText) {
+                guard let content, content.localizedCaseInsensitiveContains(searchText) else {
                     continue
                 }
             }
@@ -944,7 +970,7 @@ class MCPService {
         let readOnlyTools: [Tool] = [
             makeTool(
                 name: "read_file",
-                description: "Read the contents of a file. Maximum file size is 10MB.",
+                description: "Read the contents of a file. PDFs are supported (text layer extracted automatically). Maximum file size is 10MB.",
                 properties: ["file_path": prop("string", "The absolute path to the file to read")],
                 required: ["file_path"]
             ),

oAITests/MCPServicePDFTests.swift

@@ -0,0 +1,68 @@
+//
+//  MCPServicePDFTests.swift
+//  oAITests
+//
+// SPDX-License-Identifier: PolyForm-Noncommercial-1.0.0
+// Copyright (C) 2026 Rune Olsen
+
+import Testing
+import Foundation
+import CoreGraphics
+import AppKit
+@testable import oAI
+
+@Suite("MCPService PDF text extraction")
+struct MCPServicePDFTests {
+
+    /// Builds a real, well-formed single-page PDF (via CoreGraphics) containing the given text,
+    /// writes it to a temp file, and returns the path. Using CoreGraphics rather than hand-rolled
+    /// PDF bytes avoids flakiness from malformed xref tables etc.
+    private func makeTestPDF(text: String) throws -> String {
+        let path = FileManager.default.temporaryDirectory
+            .appendingPathComponent(UUID().uuidString)
+            .appendingPathExtension("pdf")
+
+        let pdfData = NSMutableData()
+        let consumer = CGDataConsumer(data: pdfData as CFMutableData)!
+        var mediaBox = CGRect(x: 0, y: 0, width: 200, height: 200)
+        let context = CGContext(consumer: consumer, mediaBox: &mediaBox, nil)!
+
+        context.beginPDFPage(nil)
+        let attributedString = NSAttributedString(string: text, attributes: [.font: NSFont.systemFont(ofSize: 24)])
+        let line = CTLineCreateWithAttributedString(attributedString)
+        context.textPosition = CGPoint(x: 10, y: 100)
+        CTLineDraw(line, context)
+        context.endPDFPage()
+        context.closePDF()
+
+        try (pdfData as Data).write(to: path)
+        return path.path
+    }
+
+    @Test("Extracts text from a real PDF's text layer")
+    func extractsTextFromRealPDF() throws {
+        let path = try makeTestPDF(text: "Hello World")
+        defer { try? FileManager.default.removeItem(atPath: path) }
+
+        let extracted = MCPService.shared.extractPDFText(atPath: path)
+        #expect(extracted?.contains("Hello World") == true)
+    }
+
+    @Test("Returns nil for a nonexistent path")
+    func returnsNilForMissingFile() {
+        let extracted = MCPService.shared.extractPDFText(atPath: "/tmp/definitely-does-not-exist-\(UUID().uuidString).pdf")
+        #expect(extracted == nil)
+    }
+
+    @Test("Returns nil for a file that isn't a valid PDF")
+    func returnsNilForGarbageBytes() throws {
+        let path = FileManager.default.temporaryDirectory
+            .appendingPathComponent(UUID().uuidString)
+            .appendingPathExtension("pdf")
+        try Data("not a real pdf".utf8).write(to: path)
+        defer { try? FileManager.default.removeItem(at: path) }
+
+        let extracted = MCPService.shared.extractPDFText(atPath: path.path)
+        #expect(extracted == nil)
+    }
+}