leash-poc

leash-poc: MCP tool-poisoning, contained by an independent pol...

Flowise - Remote Code Execution

Flowise 3.0.5 contains a remote code execution vulnerability caused by unsafe evaluation of user input in the CustomMCP node's convertToValidJSONString function, letting remote attackers execute arbitrary code with full Node.js privileges, exploit requires user input to be processed by the...

mcp-atlassian < 0.17.0 - Server-Side Request Forgery

MCP Atlassian 0.17.0 contains a server-side request forgery caused by improper validation of custom HTTP headers in the HTTP middleware, letting unauthenticated attackers force outbound requests to arbitrary URLs, exploit requires access to the mcp-atlassian HTTP endpoint. id: CVE-2026-27826 info...

MCP Inspector < 0.14.0 UnauthenticatedRemote Code Execution

The MCP inspector is a developer tool for testing and debugging MCP servers. Versions of MCP Inspector below 0.14.1 are vulnerable to remote code execution due to lack of authentication between the Inspector client and proxy, allowing unauthenticated requests to launch MCP commands over stdio. id...

Nginx UI - Broken Access Control

Network attackers can fully control nginx service, including config modification and service restart, leading to complete service takeover. id: CVE-2026-33032 info: name: Nginx UI - Broken Access Control author: DhiyaneshDk severity: critical description: | Network attackers can fully control ngi...

9Router <= 0.4.36 - Unauthenticated RCE

9Router = 0.4.36 middleware only guards 8 explicitly listed routes, leaving /api/cli-tools/ and /api/mcp/ entirely unauthenticated. An attacker can POST to /api/cli-tools/cowork-settings to register a custom MCP plugin with attacker-controlled command and args stored verbatim into globalThis, the...

LiteLLM 1.18.10 - Command Injection

LiteLLM 1.18.10 contains a remote code execution caused by lack of validation of arbitrary command and args in MCP server creation, letting attackers execute OS commands remotely, exploit requires crafted JSON configuration. id: CVE-2026-30623 info: name: LiteLLM 1.18.10 - Command Injection autho...

LiteLLM - Command Injection

A critical unauthenticated remote code execution vulnerability exists in LiteLLM due to improper input handling in the MCP stdio test endpoint. An attacker can send a specially crafted request to the /mcp-rest/test/connection endpoint with controlled parameters, resulting in arbitrary command...

EUVD-2026-64268

SiYuan versions before v3.8.0 contain an incomplete path blocklist in the MCP file tool that fails to restrict access to sensitive workspace files protected by the HTTP API. Authenticated administrators can read plaintext publish-mode passwords from data/.siyuan/publishAccess.json and access othe...

CVE-2026-60083 SiYuan before v3.8.0 Incomplete Path Blocklist via MCP file tool

SiYuan versions before v3.8.0 contain an incomplete path blocklist in the MCP file tool that fails to restrict access to sensitive workspace files protected by the HTTP API. Authenticated administrators can read plaintext publish-mode passwords from data/.siyuan/publishAccess.json and access othe...

CVE-2026-60083

SiYuan versions before v3.8.0 have an incomplete path blocklist in the MCP file tool that fails to restrict access to sensitive workspace files normally protected by the HTTP API. An authenticated administrator can read plaintext publish-mode passwords from data/.siyuan/publishAccess.json and acc...

CVE-2026-59809 SiYuan before v3.8.0 Secret Exfiltration via http_request URL

SiYuan before v3.8.0 interpolates secret placeholders into the destination URL parameter of the httprequest MCP tool, allowing attackers to exfiltrate stored secrets. An MCP client can craft a request with an attacker-controlled URL containing secret placeholders to send plaintext secret values t...

CVE-2026-59809

SiYuan before v3.8.0 is affected by a secret exfiltration vulnerability in the http_request MCP tool . The tool interpolates secret placeholders into the destination URL parameter, meaning an MCP client can craft a request with an attacker-controlled URL containing secret placeholders to send pla...

EUVD-2026-64267

SiYuan before v3.8.0 interpolates secret placeholders into the destination URL parameter of the httprequest MCP tool, allowing attackers to exfiltrate stored secrets. An MCP client can craft a request with an attacker-controlled URL containing secret placeholders to send plaintext secret values t...

CVE-2026-53509 @aborruso/ckan-mcp-server: SSRF via base_url allows access to internal networks (Potential fix bypass of CVE-2026-33060)

CKAN MCP Server is a tool for querying CKAN open data portals. A known vulnerability CVE-2026-33060 indicated tools including ckanpackagesearch and sparqlquery that accept a baseurl parameter had the risk of making HTTP requests to arbitrary endpoints without restriction. A fix was applied to...

CVE-2026-53509

@aborruso/ckan-mcp-server (an MCP server for querying CKAN open data portals) is vulnerable to SSRF prior to version 0.4.106 . The validateServerUrl hostname check in src/utils/http.ts only blocked the literal string localhost, but hostname aliases such as ip6-localhost and ip6-loopback (defined ...

CVE-2026-53509 @aborruso/ckan-mcp-server: SSRF via base_url allows access to internal networks (Potential fix bypass of CVE-2026-33060)

CKAN MCP Server is a tool for querying CKAN open data portals. A known vulnerability CVE-2026-33060 indicated tools including ckanpackagesearch and sparqlquery that accept a baseurl parameter had the risk of making HTTP requests to arbitrary endpoints without restriction. A fix was applied to...

CVE-2026-62674 Omnigent: Shared Agent Bundle Overwrite Leads to Authenticated Runner RCE

Omnigent is an open-source AI agent framework and meta-harness for orchestrating coding agents. Prior to 0.3.0, PUT /sessions/sessionid/agent checks LEVELEDIT permission for a session but does not reject a bound shared or template agent whose agent.sessionid is None. An authenticated user with ed...

EUVD-2026-64016

Omnigent is an open-source AI agent framework and meta-harness for orchestrating coding agents. Prior to 0.3.0, PUT /sessions/sessionid/agent checks LEVELEDIT permission for a session but does not reject a bound shared or template agent whose agent.sessionid is None. An authenticated user with ed...

CVE-2026-62674

Omnigent (open-source AI agent framework) prior to 0.3.0 is vulnerable to an authenticated Remote Code Execution via the PUT /sessions/{session_id}/agent endpoint. The endpoint validates LEVEL_EDIT permission for the session but fails to reject a bound shared or template agent whose agent.session...