Independent analysis by NHI Mgmt GroupSource article: Oak, “How to Govern MCP Servers in an Enterprise” (November 11, 2026)

TL;DR: Oak warns that MCP server sprawl turns shared, long-lived tokens into unowned machine identities that outlive the work they support, so governance has to track each server's owner, scope, usage and retirement date. Treating servers as identities collapses the gap between access creation and accountability, which is where most exposure accumulates.


At a glance

What this is: NHI Mgmt Group's analysis, drawing on Oak's blog post: MCP server sprawl makes shared tokens behave like unmanaged machine identities, so ownership, scope and retirement become the real control set.

Why it matters: IAM and NHI teams need to govern MCP servers as identities because unowned, long-lived tokens create standing access that is difficult to attribute, review or revoke cleanly.

By the numbers:

👉 Read Oak's analysis of MCP server credential sprawl and governance


Context

MCP server governance starts with a simple problem: an MCP server is not just a connector, it is an identity boundary that holds credentials and reaches downstream systems on behalf of an agent. When those credentials are long-lived, shared or created outside registration, the organisation loses the ability to explain who can act, why access exists and when it should end.

That is why the question is governance, not only security hardening. Treat each server as a named machine identity with a lifecycle, then attach ownership, scope, usage review and expiry to the token set behind it.


Key questions

Q: How should teams inventory MCP servers and their credentials?

A: Teams should inventory every MCP server, the downstream systems it reaches and the credentials it uses for each connection. That inventory should be the authoritative record for ownership, scope, review and retirement, because access that is not found cannot be governed or revoked with confidence.

Q: Why do shared tokens make MCP server access hard to govern?

A: Shared tokens erase attribution because many people and systems use the same credential, so logs show the token rather than the actual actor. That makes ownership, approval, scope review and revocation harder, especially when an MCP server token is reused across tools or teams. A unique identity per caller, short-lived tokens and central secret rotation improve governability.

Q: What is the difference between MCP server access and service account sprawl?

A: MCP server access is the permission path an agent or tool uses to call a server, usually through scoped tokens or OAuth grants. Service account sprawl is the uncontrolled growth of machine identities. The first is an access pattern, the second is an identity management problem where many accounts act on behalf of users without clear ownership.

Q: What should happen to MCP server tokens when a project ends?

A: MCP server tokens should be revoked, rotated and removed from all vaults, configuration stores and environment variables when a project ends. Any standing access tied to the server should be closed so the token cannot keep calling tools or data after the work is over. If the token was shared, every dependent secret should be reissued.


Technical breakdown

Why MCP servers behave like machine identities

An MCP server sits between an AI application and one or more downstream systems, and it executes requests using credentials it holds. That makes the server the real actor from an access-governance perspective, because the token determines what the agent can reach. If several users or agents share the same server credential, attribution collapses and the server becomes a standing access path rather than a controlled identity. See the NHI Authentication Guide.

Practical implication: Classify each MCP server and its tokens as governed identities, not informal integration details.

How credential sprawl emerges across MCP servers

Credential sprawl appears when each new server brings its own downstream tokens faster than the organisation can inventory, scope and retire them. The risk grows when builders stand up servers with personal or team credentials, then leave those tokens in place as the server proliferates. In that pattern, the number of entitlements expands even when nobody can account for the exact access paths they created. See the MCP Gateway and MCP Server Governance Guide.

Practical implication: Build a complete server-to-token inventory before any review or rotation programme can be trusted.

Why token passthrough creates borrowed reach

Token passthrough lets a server use a caller's credentials or a shared credential model that hides the effective reach behind the integration. The result is borrowed access, where the user's declared rights do not match the server's real access to downstream data and tools. That mismatch breaks the basic governance assumption that access can be certified by reviewing the originating user alone. See the MCP Security Guide.

Practical implication: Separate user intent from server entitlement so delegated reach can be reviewed on its own.


Threat narrative

Attacker objective: An attacker or careless insider aims to preserve undetected access to downstream systems through unowned server credentials.

  1. Entry occurs when teams spin up MCP servers outside a governed registration path and attach existing credentials to them.
  2. Credential access expands as each server accumulates long-lived downstream tokens that are shared across users or agents.
  3. Impact follows when unowned servers keep valid access after the builder moves on, leaving standing reach to production systems.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.


NHI Mgmt Group analysis

Shared MCP tokens are really entitlement bundles, not mere setup artefacts. When a server fronts multiple downstream systems, each token becomes a separate access decision that must be owned, scoped and retired. That is why server proliferation creates governance debt even before any attack occurs. Practitioners should treat each token as a distinct entitlement with a lifecycle, not a convenience credential.

Unowned MCP servers represent a lifecycle failure, not a visibility problem. The core gap is that the server exists longer than the accountability chain behind it. Once the builder leaves or the project ends, standing access persists unless ownership is reassigned and revocation is enforced. The governance model must assume that server creation is easy and server retirement is where control is won or lost.

Borrowed reach is the named concept that matters here. It describes the mismatch between a user's intended access and the broader reach granted through a server's token set. That mismatch is especially dangerous when agents act on behalf of users and inherit whatever the server can touch. Practitioners should audit effective access, not just declared entitlements.

Token passthrough turns attribution into guesswork. If multiple actors share the same downstream credential, the organisation cannot prove which agent or person triggered a sensitive action. That breaks review, accountability and incident reconstruction at the same time. The practical conclusion is simple: governance must preserve actor-level traceability at the server layer, not only at the user layer.

Third-party MCP servers become a governance multiplier when their tokens are standing and opaque. External integrations are not just another supplier risk when they hold long-lived access into production systems. They combine third-party exposure with internal attribution loss, which makes them harder to review than ordinary service accounts. Practitioners should place third-party server access in the highest review tier until the token set is fully explained.

From our research library:

What this signals

Borrowed reach is the control problem enterprises need to name. An MCP server can legitimately act for a user while still holding broader downstream access than the user should ever inherit. That means entitlement reviews have to inspect the server's effective reach, not just the requesting identity, because the risk sits in the delegation chain.

MCP governance will increasingly move from simple inventory to lifecycle proof. The organisations that reduce exposure will be the ones that can show who owns each server, when its tokens expire and what usage justified them, rather than relying on informal team knowledge.

Access review cadences built for human accounts will not be enough on their own if servers are created faster than reviewers can track them. The programme has to shift toward issuance-time control and retirement-time revocation for machine identities that front multiple systems.


For practitioners

  • Inventory every MCP server and downstream token Record every server, the systems it fronts and the credential used for each downstream connection. Treat any server outside the inventory as an ungoverned identity until it is registered.
  • Assign an accountable owner to each server Name a person or team responsible for each server's credentials, scope and retirement. Reassign ownership during role changes and offboarding so no server remains without a decision-maker.
  • Scope one token per server per system Avoid shared credentials across multiple users or tools. Set the smallest practical scope for each downstream system and separate read, write and admin access so excess reach is visible.
  • Review actual usage against granted access Compare what each server is allowed to call with what it actually calls in production. Remove tools and permissions that are never used, because unused access is standing access.
  • Set expiry and retirement revocation on every token Give each credential an end date, rotate persistent secrets and revoke them at the downstream system when the server is retired. Do not assume shutting the server down removes every live token.

Key takeaways

  • MCP servers should be treated as machine identities because the tokens they hold determine real downstream access.
  • Credential sprawl appears when server tokens are shared, long-lived and created outside a governed registration path.
  • The practical control set is lifecycle-based: inventory, ownership, scope, usage review, expiry and retirement.

Related NHI Mgmt Group resources: MCP Security Guide · NHI Authentication Guide · MCP Gateway and MCP Server Governance Guide · AI Agent Observability, Audit and Incident Response Guide


Standards & Framework Alignment

This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.

OWASP Non-Human Identity Top 10 and MITRE ATT&CK address the attack and risk surface, while NIST SP 800-53 Rev 5 and NIST CSF 2.0 set the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
OWASP Non-Human Identity Top 10NHI-05 — Overprivileged NHIMCP servers accumulate excess reach through shared and broadly scoped downstream tokens.
NHI-07 — Long-Lived SecretsThe article centres on static tokens that remain valid long after their original task ends.
Recommendation — Apply NHI-05 to reduce server entitlements to the minimum downstream access each server actually needs. Enforce rotation and expiry for server tokens so credentials do not outlive their intended use.
NIST SP 800-53 Rev 5IA-5 — Authenticator ManagementMCP server token lifecycle maps directly to authenticator issuance, rotation and revocation.
Recommendation — Use IA-5 to control issuance, rotation and revocation of every server credential.
NIST CSF 2.0PR.AA-05 — Access Permissions, Entitlements and AuthorizationsServer ownership, scope and usage review are all entitlement-governance problems in this post.
Recommendation — Apply PR.AA-05 to verify that each server's permissions are justified, reviewable and revocable.
MITRE ATT&CKTA0006;TA0008 — Credential Access; Lateral MovementStanding server tokens provide the access path and movement opportunities the article warns about.
Recommendation — Map exposed server tokens to credential-access and lateral-movement risk in your detection and response plans.

Key terms

  • MCP Server: An MCP server is a tool endpoint that connects an AI agent to external systems and data sources through Model Context Protocol. Because it extends what the agent can reach, it becomes part of the identity and access surface and must be reviewed like any other privileged connector.
  • Model Context Protocol: Model Context Protocol is an open protocol that lets AI agents connect to tools and data sources. It expands what an agent can reach, so governance has to cover not only the model and its prompts, but also every system that can receive or return agent-driven data.
  • Machine Identity: The digital identity of a machine, device, or workload, such as a server, container, or VM, used to authenticate it within a network. Sometimes used interchangeably with NHI, though NHI is the broader category.
  • Credential Sprawl: Credential sprawl is the uncontrolled accumulation of machine secrets, keys, and tokens across systems, teams, and environments. It usually starts with a single use case and ends with overlapping permissions, unclear ownership, and a larger attack surface than the organisation expected.

What's in the full article

Oak's full blog post covers the operational detail this post intentionally leaves for the source:

  • Step-by-step governance workflow for registering MCP servers across SaaS, homegrown and shadow environments
  • Examples of how to map owners, scopes and end dates to individual server credentials
  • A practical control table for distinguishing review, rotation and retirement actions
  • The article's own framing for how to present MCP server risk to auditors and board stakeholders

👉 The full Oak post covers the six governance controls for owning, scoping and retiring MCP server credentials.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.
NHIMG Editorial Note
Published by the NHIMG editorial team on November 16, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org