Summary
MCP Salesforce Connector exposed internal attributes in a way that allowed unauthorized disclosure of the Salesforce authentication token.
When an attacker can trigger arbitrary attribute access on the connector server, they may retrieve the auth token and reuse it to interact with Salesforce APIs as the application.
Affected context
This vulnerability affects deployments of MCP Salesforce Connector used as a Model Context Protocol (MCP) server for Salesforce integration.
It becomes relevant in environments where:
- The MCP server is reachable by untrusted or semi-trusted callers (e.g., tools, plugins, or agents)
- The connector exposes a mechanism to request or resolve attributes dynamically
- Internal authentication details (such as the Salesforce auth token) are stored in process memory or object attributes
Exploit preconditions
- Attack surface: MCP server interface of the Salesforce Connector
- Required access: Ability to send requests to the connector and trigger attribute access (e.g., via tools, extensions, or automated agents integrated with MCP)
- Trigger point: Requests that can specify or enumerate attributes of internal objects or configuration
- Why exploitation is possible: The implementation does not sufficiently restrict which attributes can be accessed, allowing exposure of sensitive fields such as authentication tokens that should remain internal-only
Why this happens
The underlying issue is an information exposure pattern where internal object attributes are made externally accessible without proper filtering or access control.
This maps to CWE-200: Exposure of Sensitive Information to an Unauthorized Actor, as the server leaks a high-value secret (the Salesforce auth token) through a generic attribute access interface.
Instead of whitelisting safe properties or encapsulating secrets behind dedicated access layers, the connector allowed arbitrary attribute resolution that included sensitive authentication data.
Potential impact
- Disclosure of the Salesforce authentication token used by the connector
- Unauthorized use of the leaked token to call Salesforce APIs
- Impersonation of the application or service account associated with the connector
- Access to sensitive Salesforce data (e.g., customer records, opportunities, reports) depending on assigned permissions
- Potential modification or deletion of Salesforce data within the token's privilege scope
General mitigation guidance
Mitigation typically focuses on restricting attribute exposure and hardening secret handling:
- Introduce explicit whitelisting of attributes that can be queried or exposed through MCP interfaces, denying access to all others by default
- Separate sensitive configuration (like tokens, client secrets, and keys) from objects exposed to callers, encapsulating them in secure storage or isolated components
- Implement least-privilege design for the Salesforce auth token, restricting its permissions to only necessary operations
- Treat all inbound attribute or field selection requests as untrusted input, validating them against a schema or allowed set
- Regularly rotate Salesforce tokens and secrets so that any unintended disclosure has limited lifetime impact
How teams detect issues like this
High-level detection approaches can include:
- Static analysis and manual code review to flag patterns where dynamic attribute access or reflection returns object internals without filtering
- Searching for interfaces that serialize or introspect objects directly (e.g., generic
__dict__exposure, automatic field dumpers) and verifying they exclude secrets - Logging and monitoring sensitive-data access paths, including when tokens or credentials are read or serialized for outbound responses
- Implementing development policies that require explicit marking or annotation of sensitive fields and enforce blocking their exposure via tooling or linting rules

