Token passthrough sends the inbound identity token to the downstream server unchanged, so both layers must trust the same issuer and audience. Token forwarding passes custom headers through the gateway to support a separate downstream auth system. Passthrough can preserve end to end identity, while forwarding should be treated as an exception path because it weakens centralized control.
How token passthrough changes the trust boundary
Token passthrough keeps one identity token intact from client to downstream MCP server. That makes the gateway part of a narrow trust chain: the downstream service must accept the same issuer, audience and token semantics the client used, and the gateway must not alter the token in transit. The benefit is consistent end to end identity; the trade-off is tighter coupling between gateway design and downstream validation rules.
In practice, passthrough is strongest when the downstream server is meant to evaluate the original caller directly rather than a gateway-minted identity. That keeps authorization decisions closer to the resource and avoids inventing a second identity layer. It also means the gateway cannot “fix up” a weak downstream trust model, so the downstream server still has to validate token audience, issuer and expiry correctly.
When a team wants a deeper governance view of the gateway patterns around approval, allowlists and audience-bound tokens, the MCP Gateway and MCP Server Governance Guide is the natural companion.
Why token forwarding is an exception path
Token forwarding does not preserve the original token as the control plane for downstream access. Instead, it passes custom headers through the gateway so a separate downstream auth system can make its own decision. That can be useful for legacy integrations or specialised downstream enforcement, but it weakens the gateway’s ability to centralise policy, normalise headers and enforce a single security model.
The key practical difference is that forwarding shifts trust from one auditable token flow to a hybrid arrangement. The gateway becomes a carrier of auth material, not the authoritative decision point, so the security posture depends on the downstream system’s parsing, header handling and policy logic. In mixed estates, that usually increases operational complexity and makes failures easier to miss.
The broader MCP security model, including the OAuth-based authorisation pattern and token passthrough guidance, is covered in MCP Security Guide, while the protocol-side authorization model is defined in Model Context Protocol: Authorization specification.
When passthrough and forwarding lead to different failure modes
Passthrough fails when the downstream server cannot safely trust the original token context, such as when audience expectations differ or the server assumes a gateway-specific token shape. Forwarding fails when teams assume the gateway has preserved policy security, but the downstream system actually depends on custom headers that are easy to mishandle, strip or duplicate. Both patterns can create confusion if the implementation does not make the trust boundary explicit.
Forwarding also increases the chance of inconsistent behaviour across services. One server may accept a forwarded header while another ignores it, so the same request path can behave differently depending on where it lands. That is why forwarding should be treated as a controlled exception, not a default architecture. In a gateway estate, the important question is not only whether the request works, but whether the authorization path remains understandable, testable and enforceable across every downstream server.
The MCP gateway control implications are easier to reason about if you compare them with gateway selection and enforcement trade-offs in the MCP Gateway Buyers Guide.
Risk and Threat Considerations
Token forwarding expands the attack surface because security now depends on both the gateway and the downstream auth system. If forwarded headers are not tightly validated, an attacker or misconfigured client can create ambiguous trust states, send conflicting identity material, or exploit differences between what the gateway thinks it passed and what the downstream service actually accepts.
Failure mechanism: The gateway forwards identity-related headers into a second authorization system without a single, enforced token contract, allowing header confusion, policy bypass or accidental overtrust in downstream parsing.
Impact: A request can be accepted under the wrong identity context, weakening centralized control, increasing the chance of unauthorized access, and making incident investigation harder because the effective auth decision is split across two layers.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP API Security Top 10 addresses the attack and risk surface, while NIST SP 800-53 Rev 5, NIST Zero Trust (SP 800-207) and OWASP ASVS set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP API Security Top 10 | API2 — Broken Authentication | Token passthrough and forwarding change API auth trust boundaries. |
| Recommendation — Validate the API auth boundary and reject ambiguous downstream token handling. | ||
| NIST SP 800-53 Rev 5 | IA-5 — Authenticator Management | Token handling here is about lifecycle and protection of authentication material. |
| IA-9 — Service Identification and Authentication | MCP gateway-to-server flows are service-to-service authentication decisions. | |
| AC-6 — Least Privilege | Forwarding can widen access paths if downstream trust is broader than needed. | |
| Recommendation — Manage token handling so forwarded or passthrough credentials stay controlled and auditable. Authenticate services directly when the downstream server must trust the caller identity. Restrict downstream access to the minimum privileges required for each MCP flow. | ||
| NIST Zero Trust (SP 800-207) | Zero Trust Architecture | The question is fundamentally about preserving or breaking explicit trust boundaries. |
| Recommendation — Enforce explicit verification at each hop instead of assuming gateway trust carries downstream. | ||
| OWASP ASVS | V10 — OAuth and OIDC | The answer depends on OAuth-style token audience and delegation semantics. |
| Recommendation — Verify audience, issuer and delegation handling before relying on token passthrough. | ||
Practitioner Guidance
Decision rule: Use passthrough when the downstream service is designed to validate the original token directly and the audience, issuer and expiry model are already agreed. Use forwarding only when you have a documented downstream auth requirement that cannot be met with passthrough, and treat that path as an exception that needs extra review.
What to verify: Confirm that the downstream server validates the exact token semantics it receives, that forwarded headers are explicitly allowed rather than implicitly trusted, and that there is no hidden dependency on gateway-specific behaviour. If the auth model cannot be described in one clear request flow, it is probably too loose.
Practitioner takeaway: Passthrough preserves a single, testable identity chain, while forwarding trades that clarity for compatibility, so the security decision should be made on the strength of the downstream trust model, not on convenience.
Related resources from NHI Mgmt Group
- How should organizations prioritize security in their MCP implementations?
- What breaks when token passthrough is allowed in MCP?
- Why does token passthrough create so much risk in MCP integrations?
- What breaks when an MCP gateway treats failed local authentication as a valid upstream passthrough case?
Deepen Your Knowledge
Free weekly newsletter
Subscribe to the NHI & AI Identity Journal
The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.
Bonus 33% off our NHI Course when you subscribe.
Reviewed and updated by the NHIMG editorial team on October 31, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org