Skip to content
MCP URL Elicitation Isn't Login: Separating Three Authorization Boundaries

MCP URL Elicitation Isn't Login: Separating Three Authorization Boundaries

The article explains that MCP's URL elicitation moves sensitive interactions outside the client but is not a login mechanism for the MCP client itself; it handles third-party authorization for the server, requiring clear separation of the user's MCP client token, the server's third-party token, and task state, with proper binding and cleanup.

Architecture Development Notes
Architecture Development Notes
Architecture Development Notes

Introduction: Two Elicitation Modes, Different Security Boundaries

MCP's elicitation mechanism lets a server request additional information from the user during a tool call. The draft specification defines two modes: form and URL. Though they appear as two UI options, they draw fundamentally different security boundaries. The form mode passes structured data to the client; the URL mode redirects the user to an address outside the client to complete the interaction.

Treating URL elicitation as "letting the MCP client log in for the server" conflates two distinct authorization relationships and risks pulling third-party access tokens into the model client.

Form Mode Is for Ordinary Input

The form mode can carry a restricted JSON Schema, allowing the client to generate a form, validate input, and let the user edit before submitting. It suits collecting names, filter criteria, or other structured data that are not credentials.

The specification explicitly forbids using form to request passwords, API keys, access tokens, or payment credentials. Because the client sees and relays the form content, any such data entering the protocol via form has already crossed a boundary it should not cross.

The client should also tell the user which server is requesting the information and provide clear reject and cancel actions. The security boundary depends not only on the server's field choices but also on the client making the request's origin and purpose visible to the user.

URL Mode Moves Sensitive Interaction Out of the MCP Client

A URL mode request contains a target address and a description. The client must clearly display the destination domain and obtain user consent before redirecting. The user then completes input, payment, or OAuth authorization on the external page.

The critical point: URL mode is not for authorizing the MCP client itself. The bearer token the MCP client uses to access the MCP server is not replaced by URL elicitation. URL mode handles the relationship where "the server needs to access a third-party service on the user's behalf."

For example, an MCP server that calls an enterprise ticketing system can generate its own third-party OAuth authorization URL, guide the user to authorize on the ticketing system's page, and then store the resulting third-party token on the server side. Subsequent tool calls use that token to access the ticketing system.

The third-party token must not be handed to the client first and then forwarded back to the server. Doing so expands the exposure surface in logs and model context and turns the MCP client into an unnecessary credential relay.

Authorization Completion Does Not Mean the Request Succeeded

The URL elicitation response only expresses whether the user agreed to open and participate in the external interaction. An accept response does not mean the third-party OAuth flow has finished, nor that the server now holds a usable token.

The server must check its own state on the next retry: has authorization completed? Which user does the token belong to? Is it still valid? If the external process is not yet done, the server should ask the user to continue or return an explicit waiting state, rather than treating a single accept as business success.

Therefore, the URL itself must not carry user credentials or sensitive personal information, and it must not be a pre-authorized link that grants direct access to protected resources just by being opened. The server must also bind the elicitation request securely to the user's identity to prevent one user's completed authorization from being reused by another user's task.

Design MCP Tools by Drawing Three Lines

Tools that require third-party authorization must distinguish at least three identities: the user invoking MCP, the MCP client, and the third-party service providing the actual resource. The MCP client's token is for accessing the MCP server; the token obtained via URL elicitation is for the server to access the third-party service; the two must never be interchanged.

In implementation, verify four things:

Are sensitive fields mistakenly placed in a form?

Is the target domain shown before redirect?

Does the server bind the third-party token to the specific user and task?

Are pending authorization states cleaned up on cancel or expiry?

The value of URL elicitation is not adding a login button, but placing the most sensitive interaction back inside the service boundary that actually requires the credentials. The MCP server remains responsible for state binding and token management; the client is responsible for letting the user know what information they are about to give to whom.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

MCPtoken managementOAuthauthorization boundariesform elicitationthird-party authorizationURL elicitation
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.