Description

MCP Documentation Server is a local-first document management and semantic search server for AI coding agents. From 1.13.0 until 1.13.1, the automatically started Web UI in src/server.ts calls startWebServer in src/web-server.ts with START_WEB_UI enabled by default and WEB_PORT set to 3080. startWebServer uses app.listen(PORT) without a host, which binds the unauthenticated document-management API to all interfaces rather than localhost. A network-reachable client can invoke GET /api/documents, GET /api/documents/:id, POST /api/documents, POST /api/search-all, DELETE /api/documents/:id, and GET /api/config without credentials to enumerate and read documents, search the corpus, insert or delete documents, and tamper with the MCP assistant's knowledge base. The service must be reachable from the attacker's LAN, VM network, container bridge, VPN, or another routed network, and the issue does not provide remote code execution. This issue is fixed in 1.13.1.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "8.8",
  "pubDate": "2026-09-17T19:16:50.517Z",
  "pubdate": "2026-09-17T19:16:50.517Z",
  "executiveSummary": "The MCP Documentation Server is susceptible to an unauthenticated API access vulnerability originating from improper network interface binding in the web server configuration.\nThe vulnerability allows remote attackers on a shared network, VPN, or adjacent container bridge to interact with the document management API without authentication.\nBy default, the server binds the administrative and data-handling endpoints to all available network interfaces rather than restricting access to the local loopback (localhost).\nThis configuration error exposes sensitive functionality, including document enumeration, retrieval, deletion, and modification, as well as the ability to tamper with the AI assistant's underlying knowledge base.\nThe risk is significant for development environments or internal deployments where the host is exposed to non-isolated networks, effectively turning a local-first tool into a network-accessible service with zero-trust controls.\nExploitation does not require prior credentials or specialized privileges, only network reachability to the service port 3080.\nThis issue affects MCP Documentation Server versions 1.13.0 through 1.13.1 and is remediated in version 1.13.1.",
  "technicalDetails": "The root cause of this vulnerability lies in the implementation of the `startWebServer` function within `src/web-server.ts`. When called by `src/server.ts`, the application utilizes the Express `app.listen(PORT)` method without explicitly specifying a host interface. In Node.js, omitting the host argument defaults the server to listening on all available network interfaces (0.0.0.0), rather than restricting the service to the local loopback address (127.0.0.1).\nThe application automatically enables the Web UI by default with `START_WEB_UI` enabled and `WEB_PORT` set to 3080. Because the server binds to all interfaces, the document management API is exposed to any device with network connectivity to the host machine. This includes attackers residing on the same LAN, those within a compromised VM network, or users across a routed VPN or container bridge.\nThe attack flow involves the following steps: (1) An attacker performs network reconnaissance to identify active listeners on port 3080. (2) Upon identifying the MCP Documentation Server, the attacker crafts unauthenticated HTTP requests to the exposed API endpoints. (3) The attacker successfully interacts with the API, which performs no validation of the requester's identity. (4) The attacker invokes methods such as GET /api/documents to enumerate the corpus, POST /api/documents to inject malicious content into the knowledge base, or DELETE /api/documents/:id to disrupt the AI agent's functionality.\nThe API endpoints exposed include sensitive functions: GET /api/documents, GET /api/documents/:id, POST /api/documents, POST /api/search-all, DELETE /api/documents/:id, and GET /api/config. By accessing GET /api/config, an attacker may also gain insights into the system's internal configuration, potentially facilitating further exploitation. The lack of authentication on these routes means any network participant possesses the same functional control as an authorized local user.\nThe post-exploitation impact is characterized by full unauthorized access to the application's document repository and knowledge management systems. While this vulnerability does not inherently provide remote code execution (RCE) on the underlying operating system, it facilitates data exfiltration and integrity compromise, allowing an attacker to manipulate the context provided to the AI coding agent, leading to indirect prompt injection or corrupted assistance outputs."
}