Summary
DeepTutor contains an authorization bypass issue where low-privilege users can invoke unrestricted MCP tools when the access control function misinterprets missing grants as full allowance.
If attackers or prompt-injected content can act within a user session, they may enumerate and call sensitive tools such as filesystem, shell, or browser servers, exposing deployment resources.
Affected context
This vulnerability arises in DeepTutor deployments that use multi-user MCP tool access controls.
It appears when the allowed_mcp_tools function treats an omitted mcp_tools field in a user's grant as unrestricted access rather than denying or limiting access.
Environments where powerful tools like filesystem, shell, or browser servers are configured are particularly exposed, especially when sessions can be influenced by untrusted or prompt-injected content.
Exploit preconditions
- Attack surface: Application-level MCP tool interface within DeepTutor sessions
- Required access: Ability to authenticate as a low-privilege user or control content within their session
- Trigger point: Invocation of MCP tools when
mcp_toolsis not explicitly defined in the user's grant - Why exploitation is possible: The access-control function returns
Noneinstead of a denied/empty set, causing the system to treat missing configuration as full access to all configured MCP tools
Why this happens
The core issue is an authorization logic flaw where absence of an explicit permission set is interpreted as unrestricted access.
In deeptutor/multi_user/tool_access.py, the allowed_mcp_tools function returns None when mcp_tools is omitted from a user's grant, rather than a clearly denied result.
This maps to CWE-862: Missing Authorization, where the system fails to properly enforce authorization checks and defaults to permissive behavior when policy data is incomplete or missing.
Potential impact
- Low-privilege accounts may gain access to powerful MCP tools such as filesystem, shell, or browser servers
- Unauthorized reading or modification of application and system files
- Execution of arbitrary shell commands within the deployment environment
- Data exfiltration from local storage or connected services
- Lateral movement within the deployment through misused system-level tools
General mitigation guidance
Mitigation efforts typically focus on enforcing explicit-deny behavior and tightening authorization logic.
Common defensive patterns include:
- Treating missing or incomplete grant configurations as no access by default
- Implementing clear, centralized authorization checks that fail closed rather than open
- Separating high-risk MCP tools into stricter privilege tiers and minimizing which roles can access them
- Designing tool invocation APIs so they always validate the user's effective permissions before execution
- Applying least-privilege principles to all tool configurations, including filesystem and shell integrations
How teams detect issues like this
Organizations can detect similar authorization issues by:
- Reviewing authorization code for "fail open" patterns when configuration fields are missing or
None - Adding static analysis or lint rules to flag permission checks that rely on implicit defaults
- Performing targeted code reviews around access-control helpers and tool dispatch functions
- Logging and monitoring MCP tool invocations, then alerting on unexpected use by low-privilege roles
- Using automated security testing to exercise sessions with minimal grants and verify that sensitive tools remain inaccessible

