CVE-2026-130796: Flowise leaks MCP bearer tokens, and no patch is coming

Flowise 3.1.3 and 3.1.4 let unauthenticated attackers read MCP server bearer tokens from public chatflows and run those flows on the owner's model quota. The project is sunset, so there is no official fix.

Christopher Ort
By Christopher Ort · 4 min read
Share
A developer staring at a computer screen displaying lines of code and API endpoint documentation in a dimly lit office.
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources

Flowise versions 3.1.3 through 3.1.4 contain an information disclosure vulnerability that allows unauthenticated attackers to read MCP server bearer tokens via the GET /api/v1/public-chatflows/:id handler. Attackers can use the leaked mcpServerConfig token on the POST /api/v1/mcp/:chatflowId route to invoke chatflows, bypassing API key protection and consuming the owner’s model quota.

i10X Quick Take

Quick Take

  1. What: CVE-2026-130796 exposes MCP server bearer tokens in Flowise 3.1.3 through 3.1.4 via an unauthenticated public API endpoint.
  2. Who is affected: Self-hosted Flowise instances running versions 3.1.3 or 3.1.4 that have a chatflow marked as public and the MCP server option enabled on it.
  3. Impact: Attackers can bypass API key protection, invoke chatflows without authorization, and consume the owner’s model quota.
  4. What to do: Flowise is being sunset and no longer accepts security reports. Users must fork the repository and apply their own fixes.
  5. CVSS score: 8.2 (High)

In plain English

Flowise is an open-source tool that helps developers build AI chatbots and workflows. Versions 3.1.3 and 3.1.4 had a bug that let anyone on the internet read secret tokens without logging in. An attacker who gets one of these tokens can use it to run the chatflow as if they were the owner, which costs the owner money because it uses up their AI model credits. The Flowise project announced in July 2026 that it is shutting down and will no longer fix security problems.

What you should do: If you run Flowise yourself, you need to either stop using it or download the code and fix this bug yourself.

What happened

The vulnerability exists in the GET /api/v1/public-chatflows/:id endpoint, which is designed to return sanitized chatflow data for public use. According to the source code in the Flowise GitHub repository, the handler calls sanitizeFlowDataForPublicEndpoint but does not strip the mcpServerConfig field from the ChatFlow entity before returning it.

The ChatFlow entity stores mcpServerConfig as a plain text column holding a JSON object that includes the MCP bearer token. According to the write-up by security researcher geo-chen, the public endpoint returns the full chatflow database row and only sanitizes the flowData field, so the token and other settings such as apikeyid and apiConfig are returned verbatim. The researcher notes that both preconditions, a public chatflow (used for embedded and shared chatbots) and an enabled MCP server, are common and set independently.

An unauthenticated attacker can retrieve this token by sending a GET request to /api/v1/public-chatflows/:id for any public chatflow. The attacker can then use the token in the Authorization header when calling the POST /api/v1/mcp/:chatflowId endpoint, which validates the bearer token against the stored mcpServerConfig.token value. This allows the attacker to invoke the chatflow without a valid API key, bypassing quota controls and consuming the owner’s model credits.

The vulnerability affects only Flowise versions 3.1.3 and 3.1.4. The sources do not state whether earlier or later versions are affected.

What you should do

Flowise announced on its sunset page that it is winding down operations. The project entered a code freeze on July 29, 2026, and the GitHub repository was scheduled for public archive status on August 10, 2026, with end of life on August 31, 2026. The SECURITY.md file states that Flowise is no longer accepting new security vulnerability reports.

No official patch or fixed version is available. Users running affected versions must either discontinue use of Flowise or fork the FlowiseAI/Flowise repository on GitHub and apply their own fixes. The Apache 2.0 license permits such modifications.

To mitigate the vulnerability manually, developers should modify the sanitizeFlowDataForPublicEndpoint function or the GET /api/v1/public-chatflows/:id handler to strip the mcpServerConfig field before returning chatflow data to unauthenticated users. The researcher recommends returning only a whitelisted set of fields, as the separate public chatbot config endpoint already does. Until then, admins can disable the MCP server option on public chatflows, or stop marking chatflows as public, and regenerate MCP tokens that may already have leaked.

What it means

This vulnerability demonstrates the risk of exposing internal configuration data through public API endpoints. Bearer tokens intended for server-to-server authentication should never be included in responses to unauthenticated clients.

The shutdown of the Flowise project leaves users without official support or patches. Organizations relying on Flowise for production workloads face a choice between migrating to alternative platforms or maintaining their own forks with security fixes.

The CVSS score of 8.2 reflects the high impact of unauthorized access and quota consumption. Exploitation needs the ID of a public chatflow, which is often visible to anyone who can open the embedded or shared chatbot.

Leaky MCP endpoints keep turning up in AI tools: recent cases include an unauthenticated HTTP flaw in the Dynatrace MCP server and unauthenticated code execution in AgentScope MCP workspaces. New cases are tracked in our AI Security hub.

Sources

  • NVD entry CVE-2026-130796 (published November 17, 2026)
  • Researcher write-up by geo-chen: Unauthenticated disclosure of MCP bearer token and config secrets via GET /api/v1/public-chatflows/:id
  • Flowise sunset announcement
  • Flowise chatflows controller source code
  • Flowise ChatFlow entity schema
  • Flowise MCP server service source code
  • Flowise SECURITY.md
Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. Microsoft Teams Lets AI Agents Join as Digital Teammates With Their Own Identity
  2. Azure Cobalt 200 VMs Reach General Availability With Up to 50% More CPU Performance
  3. Microsoft Launches AI Transparency Initiative With NIST Standards Work and 40 Partners
  4. Agent 365 Gets Business Analytics and FinOps for AI as Nadella Cites 50 Million Agents
  5. Microsoft Perception Is Generally Available, With Red, Blue and Green Security Agents