---
id: "CVE-2026-54140"
kind: "vulnerability"
aliases: ["CVE-2026-54140"]
title: "OpenHuman: MCP server tool descriptions can inject instructions into agent"
severity: "high"
cvss: 8.3
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-agents", "ai-tooling", "desktop", "mcp", "openhuman"]
not_affected_if: ["no-external-mcp-servers", "trusted-mcp-servers-only"]
published_at: "2026-11-05T17:18:18+00:00"
first_seen_at: "2026-11-05T17:40:04+00:00"
updated_at: "2026-11-05T17:42:12+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-54140", "first_seen_at": "2026-11-05T17:40:04+00:00"}]
exploit_sources: []
---

# OpenHuman: MCP server tool descriptions can inject instructions into agent

OpenHuman, an open-source agent harness, passes names and descriptions from configured external MCP servers straight into the model's context without filtering. A malicious or compromised MCP server can hide instructions in those descriptions that make the agent call other tools, disclose data, or take unauthorized actions.

_Summary, check and fix written by a model from the sources; labelled as such._

## How to check

- Check the installed OpenHuman version and compare it with `0.57.13`
- Review the list of external MCP servers configured in OpenHuman and confirm each one is trusted

## Fix

- OpenHuman: move to `0.57.13` or later

## Timeline

- 2026-11-05T17:40:04+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-54140

## Documents

- [advisory] Vendor advisory: github.com https://github.com/tinyhumansai/openhuman/security/advisories/GHSA-g6j9-89xv-v752
- [fix] Patch: github.com https://github.com/tinyhumansai/openhuman/commit/00b8a669c6e6391a1be9f724bd9c8f755ee3c43e
- [fix] Release notes: github.com https://github.com/tinyhumansai/openhuman/releases/tag/v0.57.13

## From the sources

OpenHuman is an open-source agent harness available as a desktop application and Rust library. Before 0.57.13, OpenHuman places names, descriptions, titles, and parameter definitions returned by configured external MCP servers into the agent model context without neutralizing instructions or bounding the descriptive fields. The McpRemoteTool records produced by list_tools preserve this remote content, while tool filtering considers the tool name and the prompt-injection detector screens user messages rather than MCP definitions. A malicious or compromised MCP server can therefore return a poisoned description that is evaluated on every agent turn and may direct the model to invoke other available tools, disclose accessible data, or perform unauthorized actions without a separate malicious user message. The issue affects users who configure an attacker-controlled or compromised MCP server. This issue is fixed in version 0.57.13.

---

_Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai_