NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · Other MCP servers and tools

CVE-2025-5273: Markdownify MCP Server allows attackers to read arbitrary files

mediumCVSS 4.0 6.9 · 3.1 6.5GHSA-22v8-p7h2-rj7p

CVE-2025-5273 lets a crafted prompt or web page make an AI agent read arbitrary files through the get-markdown-file tool of mcp-markdownify-server. The GitHub advisory data lists 0.0.1 and earlier as vulnerable with no patched version. Snyk lists versions below 1.0.0 as affected and 1.0.0 as the fix.

Published May 29, 2025 · Analysis Nov 1, 2026

What an attacker can do

Markdownify MCP converts files and pages to Markdown for an AI assistant. Its get-markdown-file tool (src/Markdownify.ts, around line 94) is not limited to an allowed directory. Snyk's proof of concept passes a file:/// URI for /etc/passwd, which the tool reads and returns to the model.

The attacker cannot call the tool directly. They plant instructions where the agent will see them, such as hidden text on a web page. If the agent loads the page and follows the instructions, it reads a sensitive local file, such as SSH keys, .env files or cloud credentials, and can then send the contents to an attacker-controlled URL using other tools. The CVSS vectors mark user interaction as required or active, no privileges, and high confidentiality impact only.

The risk grows when the same agent also has a tool that can send data out, such as a fetch or browse tool. With a read tool that reaches any file, and an outbound channel, one poisoned page is enough to leak secrets. A public proof of concept exists.

Who is affected

The npm package mcp-markdownify-server. The GitHub advisory data lists 0.0.1 and earlier; Snyk says versions below 1.0.0. The advisory text says all versions are vulnerable. Check what you use:

npm ls -g mcp-markdownify-server

If the server is launched from a clone of the zcaceres/markdownify-mcp repository, check the commit in the checkout. Look for the server in your MCP client config.

mcp-markdownify-server

Ecosystem
npm
Affected
<= 0.0.1
Fixed in
No patched version yet
Check
npm ls mcp-markdownify-server

How to fix it

Snyk names version 1.0.0 as the fix, and links the commit 3a6b202. The GitHub advisory data lists no patched version, so confirm the version that you install:

npm install -g mcp-markdownify-server@latest

If you cannot confirm a fixed build, remove the server from your MCP client or turn off its get-markdown-file tool. Do not let an agent with this server read untrusted pages and also hold a tool that can send data out. Run the server as a user with no access to secrets.

How to tell if you were hit

Review agent transcripts and MCP logs for get-markdown-file calls with file:/// URIs or paths outside the project, especially ones that follow the agent loading a web page. Look for reads of /etc/passwd, ~/.ssh and .env files followed by an outbound request. Beam records agent file access and MCP tool calls, and its rules can flag reads of sensitive files, which fits this path.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

All versions of the package mcp-markdownify-server are vulnerable to Files or Directories Accessible to External Parties via the get-markdown-file tool. An attacker can craft a prompt that, once accessed by the MCP host, will allow it to read arbitrary files from the host running the server.

Questions

Is there a fixed version of mcp-markdownify-server?

Snyk says 1.0.0. The GitHub advisory data lists no patched version and 0.0.1 as the last vulnerable one, so check which version you install.

How is this different from CVE-2025-5276?

This one reads local files through get-markdown-file. The other is an SSRF through Markdownify.get(), reachable from the webpage, Bing search and YouTube tools.

Does the attacker need access to my machine?

No. They need the agent to process content they control, such as a web page with hidden instructions.

Is a proof of concept public?

Snyk reports one. We found no report of exploitation in the wild.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More Other MCP servers and tools vulnerabilities

All Other MCP servers and tools entries