A resource is a read-only MCP object identified by a URI and fetched when the model needs supporting context. Resources are useful for exposing data without granting action rights, but they still carry governance implications if the data can influence later tool selection or reveal sensitive state.
What a resource is in MCP
In Model Context Protocol, a resource is a read-only object identified by a URI and retrieved when the model needs supporting context. That design keeps data exposure separate from action rights: a resource can inform the model without letting it directly change systems, which is one reason MCP distinguishes resources from tools.
That separation matters because the object can still carry operational meaning. A resource may expose configuration, state, or reference data that influences later reasoning, so the security question is not only “can it act?” but also “what can it reveal, and how might that shape downstream decisions?”
How resources differ from tools and prompts
Resources are not the same as tools. Tools are for actions, while resources are for retrieval. In practice, that means a resource should be treated as a source of context, not an execution surface, even though the model may use what it reads to choose a later tool or adjust its plan.
That distinction is important in governance reviews. If a system exposes the wrong thing as a resource, it may unintentionally expand the model’s awareness of internal state, secrets-adjacent details, environment names, or policy-relevant context. The issue is often not direct action abuse, but overexposure through read access.
Why URI-based retrieval matters
Because each resource is identified by a URI, resource design is partly an information architecture problem. Stable naming, clear scoping, and predictable retrieval help the client know what it is asking for and help operators understand what data the model can reach. Poorly governed URIs can turn a neat abstraction into an inventory and access problem.
Read-only access also does not eliminate trust concerns. A resource that returns sensitive or highly specific data can still become a security liability if the model later incorporates it into tool selection, summaries, or responses. For that reason, resource governance often sits close to content classification, access review, and environment segregation.
What good resource governance should protect
Good resource governance focuses on exposure, correctness, and least privilege for context. A resource should expose only the data needed for the model’s task, and operators should know which resources are available, which ones are sensitive, and which ones are safe to surface in shared or delegated workflows.
That is especially important where the resource may describe live system state, tenant-specific data, or operational metadata. Even without write capability, a resource can leak enough context to create privacy, confidentiality, or decision-shaping risk, so governance must track both the data itself and the downstream influence it can exert.
Risk and Threat Considerations
Resources create risk when read access becomes a back door to sensitive context. The main failure mode is not direct system manipulation, but unintended disclosure, model steering, or exposure of state that helps an attacker understand the environment or shape later interactions.
Failure mechanism: An overbroad or poorly classified resource can reveal information that the model then reuses in subsequent reasoning, prompting, or tool selection, even though the resource itself is read-only.
Impact: That can expose confidential data, increase the blast radius of prompt or workflow abuse, and create indirect control risk if the disclosed context influences privileged follow-up actions.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
NIST SP 800-53 Rev 5 provides the primary governance reference for this term.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev 5 | AC-6 — Least Privilege | Resource access should expose only the context needed by the model. |
| AU-2 — Event Logging | Resource retrieval and downstream use need traceability in audit records. | |
| SC-4 — Information in Shared Resources | Resources share data without granting action rights, so shared-data controls apply. | |
| Recommendation — Limit resource exposure to the minimum data needed for the task. Log resource access and downstream context use for review. Protect sensitive information when it is exposed through shared read paths. | ||
Practitioner Guidance
Why practitioners should care: Resource design is a governance decision, not just an implementation detail. Teams should decide which data belongs in a resource, who may expose it, and whether the content is safe for model consumption in the first place.
What to watch for: The highest-risk cases are resources that bundle operational state, sensitive metadata, or environment-specific information that was never meant to be broadly visible to model consumers. Those resources deserve the same review discipline you would apply to other read paths with security impact.
Related resources from NHI Mgmt Group
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 November 5, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org