# Reported issues for PDF Reader MCP Pod holds 24 of 38 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user. Back to [PDF Reader MCP](/mcp/pdf-reader-mcp). ## Most discussed ### [BUG] Unable to start MCP server in VSCode ## 🐛 Bug Description MCP server is not getting start itself. Error Logs: ` 2025-12-03 15:43:51.479 [warning] [server stderr] SyntaxError: The requested module '@sylphx/gust' does not provide an export named 'router' 2025-12-03 15:43:51.479 [warning] [server stderr] at ModuleJob._instantiate (node:internal/modules/esm/module_job:228:21) 2025-12-03 15:43:51.479 [warning] [server stderr] at async ModuleJob.run (node:internal/modules/esm/module_job:337:5) 2025-12-03 15:43:51.479 [warning] [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/240) · 2025-12-03 · closed · external user · 6 comments ### Struggling to access the PDF file multiple times before it suceeds ## 🐛 Bug Description MCP, struggling to access the PDF file multiple times before it succeeds ## 📋 Steps to Reproduce 1. Do this in Windows OS 2. Have a PDF file with “space” inside the filename (e.g, “Test File Bug.pdf”) 3. Open in VS Code, try to ask GitHub Copilot to read the file content 4. The AI will fail a couple of times to access the file 5. The file should be accessed, but it takes multiple tries to try to open the file. ## ✅ Expected Behavior The MCP immediately access the file, [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/235) · 2025-11-28 · closed · external user · 3 comments ### no tools After I installed，mcp server has no tools [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/186) · 2025-10-14 · closed · external user · 5 comments ### Can it allow AI to read images in PDFs? Can it extract pictures? It is not clearly mentioned in the readme. [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/180) · 2025-10-08 · closed · external user · 5 comments ### MCP error -32602: Absolute paths are not allowed Hi! Would there be any way to make this work? It's easier for the AI to access the file and the user to point to it if the user can simply copy paste the file direct link. { "sources": [ { "path": "C:\\Users\\SANDBOX\\Desktop\\sample-2.pdf" } ], "include_full_text": true, "include_metadata": true, "include_page_count": true } { "results": [ { "source": "C:\\Users\\SANDBOX\\Desktop\\sample-2.pdf", "success": false, "error": "MCP error -32600: Faile [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/136) · 2025-08-20 · closed · external user · 8 comments ### getAll() does not exist Hi! I tried to build locally but I get following error on freshly cloned repository: > @sylphlab/pdf-reader-mcp@0.3.24 build > tsc src/handlers/readPdf.ts:256:40 - error TS2339: Property 'getAll' does not exist on type 'Metadata'. 256 const metadataData = metadataObj.getAll() as PdfMetadata | undefined; ~~~~~ Found 1 error in src/handlers/readPdf.ts:256 Can you fix it please? [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/64) · 2025-05-23 · closed · external user · 8 comments ### Tool schema incompatible with Mistral API (400 error: "Invalid tool schema") Hi, Thank you for developing this project! I’m encountering a problem when running `pdf-reader-mcp` alongside the Mistral API. When the PDF Reader MCP server is enabled, any API call to Mistral that includes the registered tool schema fails with the following error: ``` API error occurred: Status 400 Content-Type application/json Body {"object":"error","message":"Invalid tool schema: {'type': 'object', 'properties': {'path': {'type': 'string', 'minLength': 1, 'description': 'Relative path t [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/63) · 2025-05-22 · closed · external user · 5 comments ### File not found error Hi, My agent is calling the tool with these args: ``` { "sources": [ { "path": "FooBar.pdf" } ], "include_full_text": true, "include_metadata": true, "include_page_count": true } ``` And getting this response: ``` { "results": [ { "source": "FooBar.pdf", "success": false, "error": "MCP error -32600: File not found at 'FooBar.pdf'." } ] } ``` And I can confirm that the PDF exists. If I run the server directly (without making a request), I [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/21) · 2025-04-15 · closed · external user · 5 comments ## Most recent ### [BUG] MCP Spec 2026-07-28 conformance failures in `server/discover` and `tools/list` result envelopes cause client initialization to fail ### 🐛 Bug Description When connecting `@sylphx/citra@5.0.0` (pure-Rust `citra-mcp-server`) to modern MCP clients (such as Gemini Antigravity CLI and MCP 2026 spec runners), the server fails to initialize and register tools. This happens due to two related protocol issues in `crates/pdf-reader-mcp-server/src/discover_compat.rs`: 1. **Schema Violations in `server/discover` (SEP-2575):** The payload returned by `discover_result_value()` is missing required fields defined in the MCP 2026-07-28 sp [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/663) · 2026-08-25 · closed · external user · 1 comment ### MCP spec conformance: 13 requirement(s) violated (via @hasmcp/mcp-spec-test) — spec 2026-07-28 When the MCP server published from this repo (`npx -y @sylphx/citra`) is tested against the newest MCP spec revision (2026-07-28) with `@hasmcp/mcp-spec-test`, 13 requirements are violated: `server/discover`'s response is missing required fields (`cacheScope`, `ttlMs`), `tools/list`/`tools/call` results are missing required envelope fields (`resultType`, `_meta.serverInfo`, cache hints), and a couple of version-negotiation edge cases cause the server process to exit rather than respond. Notably, [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/662) · 2026-08-24 · closed · external user · 1 comment ### [pdf-reader-core 3.1.1] cff-parser panic at encoding.rs:150 on CFF fonts with Custom encoding (Adobe InDesign + iTextSharp PDF) ## Summary `@sylphx/citra` v5.0.0 (native engine `pdf-reader-core` 3.1.1) **panics (explicit `panic!`)** while extracting text from a PDF that embeds CFF / Type 1C fonts with **Custom** encoding. The panic originates in the `cff-parser` Rust crate and kills the native engine, surfacing on the SDK/MCP side as a `timeout tools/call` error. ## Environment - OS: macOS (darwin) arm64 — Apple Silicon - Node.js: v22.22.2 - Package: `@sylphx/citra@5.0.0` - Native engine: `pdf-reader-core` 3.1.1 (bina [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/660) · 2026-08-22 · closed · external user · 1 comment ### Your MCP server is listed on DReview Hi — just wanted to flag that PDF Reader MCP is listed on DReview (https://dreview.co/tools/pdf-reader-mcp), a free directory of MCP servers with per-client install configs. No action needed on your end — feel free to link to the listing from your README if useful, or let me know if anything about the listing needs a correction. Thanks for building this — it's a genuinely useful server. [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/622) · 2026-08-06 · closed · external user · 1 comment ### [BUG] The pdf reader crashes on the file provided ## 🐛 Bug Description When reading the file, the MCP crashes with `panicked at adobe-cmap-parser: bad length of hexstring` ## 📋 Steps to Reproduce 1. Run any mcp operation on the file provided. ## ✅ Expected Behavior The tool will work. ## ❌ Actual Behavior The tool crashes. ## 📸 Screenshots / Logs ``` p [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/608) · 2026-08-02 · closed · external user · 1 comment ### [BUG] Crash on server/discover during Gemini Antigravity CLI MCP client initialization. ## 🐛 Bug Description The `pdf-reader-mcp` server crashes immediately during initialization when connected to the Gemini MCP client. The crash happens because the Gemini client sends a custom `server/discover` request before completing the standard JSON-RPC handshake, causing an unhandled exception and a premature stream closure (`EOF`) on the server side. ## 📋 Steps to Reproduce 1. Install `pdf-reader-mcp` and add it to the Gemini Antigravity CLi MCP plugins (`"~\.gemini\config\mcp_config.jso [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/598) · 2026-07-27 · closed · external user · 1 comment ### [BUG] OpenCode中配置后无法使用 ## 🐛 Bug Description opencode配置mcp后请求报错：Invalid schema for function 'PDF____read_pdf': {"type":"object","properties":{"path":{"type":"string","minLength":1,"description":"Relative path to the local PDF file."},"url":{"type":"string","format":"uri","description":"URL of the PDF file."},"pages":{"anyOf":[{"type":"array","items":{"type":"integer","exclusiveMinimum":true,"minimum":0},"minItems":1},{"type":"string","minLength":1}],"description":"Extract text only from specific pages (1-based) or ran [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/562) · 2026-07-23 · closed · external user · 3 comments ### read_pdf url source: SSRF and file:// local-file read bypass the path-confinement that protects the path source reported on 5 June 2026 https://github.com/SylphxAI/pdf-reader-mcp/security/advisories/GHSA-34gp-w56h-r2mv - no response so trying here: ### Summary The read_pdf tool accepts sources as either a project-relative path or a url. The path branch is correctly confined to the project root (absolute paths and ../ are rejected), but the url branch applies no validation beyond a URL-format check, so it can fetch internal/metadata HTTP endpoints (SSRF) and, via file:// URLs, read PDF files anywhere on [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/368) · 2026-07-08 · closed · external user · 2 comments ### [BUG] Security Vulnerabilities Found src/utils/pathUtils.ts SSS-03: Path Traversal and Symlink Escape — Critical Location: lines 28-35 Finding: resolvePath() uses path.resolve() and path.normalize() to process user-provided paths but never calls fs.realpathSync() to canonicalize symlinks. When --allow-dir restrictions are configured, a symlink inside an allowed directory can point outside it, bypassing the containment check in isPathAllowed(). path.resolve('allowed/link/target') passes startsWith('allowed/') but the real path may [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/279) · 2026-05-15 · closed · external user · 0 comments ### Ontheia listed as compatible client — works great with pdf-reader-mcp > Hi, > > We've been testing your MCP server with [Ontheia](https://ontheia.ai) — a self-hosted, open-source AI agent platform — and it works great out of the box. > > We've added it to our [Compatible MCP Servers](https://docs.ontheia.ai/getting-started/03_compatible-mcp-servers/) page to help our users discover it. > > If you'd like to list Ontheia as a compatible client in your README, we'd be happy about that — but no obligation at all. > > Thanks for building this! > — Wolfgang, founder of [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/276) · 2026-04-21 · closed · external user · 1 comment ### [FEATURE] limit gobal filesystem access and http download ## 💡 Feature Description It seems by default pdf-reader-mcp supports reading files from absolute and relative filesystem paths and http URLs. This means it can access without explicit permission quite a lot of content, with the security implications this has. It is understandable that some filesystem access is needed to get binary data in, but it would be great if this was possible to limit, like for the Filesystem reference MCP. Suggest having a command line option to opt out of general file [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/274) · 2026-04-19 · closed · external user · 1 comment ### [BUG] Missing OpenJPEG WASM fallback file causing image decode failures ## 🐛 Bug Description When processing PDFs with images, the OpenJPEG WASM module fails to initialize due to a missing fallback file (named 'nullopenjpeg_nowasm_fallback.js'), causing image decode failures across all pages. ## 📋 Steps to Reproduce 1. Install pdf-reader-mcp via npx or npm 2. Attempt to process a PDF containing images 3. Observe the OpenJPEG errors in the logs ## ✅ Expected Behavior Images within PDFs should decode correctly or gracefully fall back to non-WASM/Javascript implem [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/271) · 2026-04-09 · closed · external user · 1 comment ### [FEATURE] Add support for HTTP transport ## 💡 Feature Description Thanks for this awesome project! Since this MCP server is currently purely running locally, I would like to request adding support for HTTP transport (see https://modelcontextprotocol.io/specification/2025-06-18/basic/transports#streamable-http). ## 🎯 Problem Statement By adding HTTP transport as an (configurable) alternative, one could host this MCP server somewhere on the public internet and make it accessible by all his/her workplaces. **Current Situation:** Thi [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/255) · 2025-12-28 · closed · external user · 1 comment ### [BUG] vscode gets unresponsive First of all: very cool project! I really like the idea. ## 🐛 Bug Description When I try summarizing a larger PDF (20+ pages), it takes forever, and the VS Code window becomes unresponsive almost immediately, even though my CPU usage is only around 20%. ## 📋 Steps to Reproduce It would be great if you could test this behavior and let me know if my machine is simply too weak to handle it! 1. Set up VS Code 2. Try summarizing larger [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/254) · 2025-12-26 · closed · external user · 2 comments ### Add CMap support for Japanese/CJK PDF text extraction Japanese text extraction fails for certain PDFs (e.g., Adobe InDesign-generated corporate reports). Only a few English words are extracted while Japanese text is missing. ## Test Case ``` URL: https://colopl.co.jp/assets/pdf/ir/library/reports/COLOPL_Report_2024_print.pdf Page: 2 Expected: Full Japanese text (コロプラはGPSを活用した...) Actual: Only "GPS17Entertainment in Real LifeCOLOPL Report 202401" ``` ## Root Cause In `src/pdf/loader.ts` line 77, `getDocument()` is called without CMap parameters [Read the thread](https://github.com/SylphxAI/pdf-reader-mcp/issues/250) · 2025-12-16 · closed · outside contributor · 1 comment ### [DOCS] Modelscope page outdated ## 📚 Documentation Issue The page published on Modelscope is severely outdated. The provided installation command incorrectly installs a seemingly similar legacy version instead. This makes it extra difficult for users to pinpoint the cause of problems after installation. If this page was published by your team, please consider updating it. ## 📍 Location Image