Browse / MCP and Tool Server Flaws
MCP and Tool Server Flaws
Vulnerabilities in Model Context Protocol servers, clients, plugins and agent tool integrations. 177 records. Reference: OWASP Agentic · LLM03.
- Authentication and access control 38
- Code and command execution 37
- Path traversal and file access 27
- Server-side request forgery 23
- Denial of service 14
- Data exfiltration 10
- Information exposure 8
- Other weaknesses 6
- Cross-site scripting 5
- Cross-site request forgery 3
- Sandbox escape 2
- Credentials from AI Agent Configuration 1
- AI Agent Tool Poisoning: Implementation 1
- AI Agent Tool Poisoning: Definition and Instructions 1
- Discover AI Agent Configuration: Tool Definitions 1
- Poisoned Postmark MCP Server Email Exfiltration
- Flowise has Remote Code Execution vulnerability The CustomMCP node passes the mcpServerConfig string to the Function constructor, so a crafted configuration value runs arbitrary JavaScript with Node.js privileges.
- Flowise has unsandboxed remote code execution via Custom MCP The Custom MCP feature launches attacker-chosen commands through a load-method endpoint that lacks proper authentication and role checks.
- MCP Inspector is Vulnerable to Potential Command Execution via XSS When Connecting to an Untrusted MCP Server The inspector renders a redirect URL from an untrusted MCP server without sanitising it, allowing script injection that reaches the local proxy.
- Flowise OS command remote code execution Custom MCP server configuration lets unauthenticated network users run arbitrary OS commands through the server spawn feature.
- mcp-package-docs vulnerable to command injection in several tools Unsanitised tool parameters are concatenated into a child_process.exec shell string, so shell metacharacters run extra commands on the server.
- mcp-remote exposed to OS command injection via untrusted MCP server connections A malicious MCP server returns a crafted authorization endpoint URL that mcp-remote passes into an OS command, executing commands on the client.
- MCP Python SDK vulnerability in the FastMCP Server causes validation error, leading to DoS A validation error on malformed requests in the MCP Python SDK is unhandled, so the FastMCP server returns errors until it is restarted.
- MCP Python SDK has Unhandled Exception in Streamable HTTP Transport, Leading to Denial of Service A client that deliberately triggers an error mid-session causes an unhandled closed-resource exception that crashes the MCP streamable HTTP server.
- @modelcontextprotocol/server-filesystem vulnerability allows for path validation bypass via colliding path prefix The filesystem server's allowed-directory check is a plain prefix match, so sibling directories sharing the allowed prefix pass.
- @modelcontextprotocol/server-filesystem allows for path validation bypass via prefix matching and symlink handling Allowed-directory checks in the filesystem MCP server can be defeated by prefix matching and symlinks, exposing files outside the permitted roots.
- @cyanheads/git-mcp-server vulnerable to command injection in several tools Git MCP tools pass unvalidated parameters into child_process exec shell strings, so shell metacharacters run arbitrary commands, including when a model is steered by injected log content.
- Data Exfiltration via an MCP Server used by Cursor
- Claude Code Improper Authorization via websocket connections from arbitrary origins A malicious webpage opens a websocket to the local Claude Code IDE server, which accepts it without checking origin, letting it read files and IDE state.
- MCP Inspector proxy server lacks authentication between the Inspector client and proxy The MCP Inspector proxy accepts unauthenticated requests that launch stdio commands, letting remote callers execute code on the host.
- Data Exfiltration via Remote Poisoned MCP Tool
- AgentScope arbitrary file download vulnerability in rpc_agent_client The agent RPC client's download method accepts any file path on the remote host, letting any user fetch arbitrary files from it.
- composio allows Server-Side Request Forgery (SSRF) in BROWSERTOOL Browser tool actions pass attacker-supplied URLs to the server-side fetcher, which can be pointed at local resources such as file paths to read arbitrary files.
- Composio Eval Injection Vulnerability The calculator tool passes user-supplied expressions to eval(), so a crafted math input runs arbitrary code in the server process.
- composio Server-Side Request Forgery (SSRF) vulnerability A web scraping tool endpoint fetches user-supplied URLs without restriction, letting callers reach internal services and cloud metadata.
- langflow has vulnerability in PythonCodeTool component The PythonCodeTool component in Langflow lets remote callers run arbitrary Python through the tool integration.
- @lobehub/chat vulnerable to unauthorized access to plugins The plugin gateway endpoint accepts requests without the access code that protects the chat interface, so plugins run unauthenticated.
- Arbitrary Code Execution via Scanner Bypass in aws-diagram-mcp-server exec() Namespace
- Command Injection on Amazon Q Developer CLI via malicious .amazonq/mcp.json leads to arbitrary code execution
- Second-Order XSS via javascript protocol in MCP Server Portal Apps leads to ATO
- DNS Rebinding SSRF in Burp Suite MCP Server Enables Internal Network Access via send_http1_request Tool
- use-mcp's oauth2 process uses a window.open call with untrusted mcp server provided data allowing for code execution