User Token Mapping and Chained Downstream Tokens for MCP / ToolHive

ToolHive needs to maintain a per-user mapping of tokens across multiple upstream services (e.g., Google, GitHub, AWS). When a user authenticates, ToolHive can chain the consent process: alongside granting access to ToolHive/MCP, the user is also prompted to authorize against one or more external IdPs. ToolHive then keeps both the internal and external tokens, refreshes them as needed, and injects or exchanges the correct downstream token when MCP servers call external APIs.

Tasks:

• [ ] Provide functionality into Toolhive proxy for users to go into a consent flow that yields both a ToolHive token and one or more external tokens (e.g., Google, GitHub).
• Provide functionality into Toolhive proxy to store and refresh external tokens securely, linked to the internal user identity. They need to be encrypted, refreshed/rotated when supported, and never logged.
• Provide functionality into Toolhive proxy to retrieve the right mapped downstream token and provide it to the MCP server when needed.
• Create an end-to-end demo showing chained login → token mapping → MCP call with downstream token injection.
• [ ] Create documentation explaining the chained consent model, token storage, and operator security practices.
• Perform a security review validating protections against token leakage, replay, mis-linked identities, and consent confusion.

Start by reviewing the ToolHive proxy's existing authentication and MCP request flow, then map the consent, storage, refresh, and downstream-token requirements. Use the end-to-end demo and security review as completion criteria, with documentation covering the consent model, storage, and operator practices.