The process by which an MCP server or proxy exposes available tools to a caller. If discovery is not filtered by principal and context, it can reveal capabilities that the identity is not actually authorised to use, creating an exposure before execution begins.
What MCP Tool Discovery Actually Means
MCP tool discovery is the step where an MCP server or proxy publishes the tools it offers so a caller can inspect what is available before invoking anything. That makes discovery a control point, not just a catalog function, because the tool list itself can expose capabilities, workflow paths, and hidden reach that should be filtered by principal and context.
In practice, discovery is where the interface between availability and authorisation becomes visible. If the server presents every tool to every caller, the caller may learn about capabilities that exist but were never meant to be shown, let alone used, by that identity.
Why Discovery Is Security-Sensitive
Tool discovery can leak more than names. Tool metadata may reveal privileged actions, sensitive integrations, environment structure, or the existence of administrative functions that an attacker or untrusted caller can use to map the system. Model Context Protocol: Authorization specification shows why discovery and authorization need to be aligned, because tokens and resource access are meant to be scoped rather than broadly replayed.
This is especially important in agentic environments, where a tool list can become an attack surface on its own. A caller that sees too much can probe too much, and an agent that is allowed to discover too much may later attempt actions outside its intended reach.
Filtering Discovery by Principal and Context
Good MCP discovery is conditional. The tools exposed to a human user, a delegated agent, a service process, or a test environment may need to differ because the right answer depends on who is asking, from where, and under what policy. If those filters are missing, the discovery response becomes an information disclosure channel even when execution is still blocked.
That means discovery should reflect the same policy logic that will govern execution, not a broader global catalog. The more an MCP server supports mixed audiences, the more important it is to avoid treating discovery as a static, one-size-fits-all list.
For readers evaluating the wider agent risk picture, OWASP Agentic AI Top 10 is useful context because it frames tool misuse and identity abuse as core agentic risks rather than edge cases.
What Safe MCP Tool Discovery Should Support
Safe discovery should help a caller understand what is relevant without exposing what is off-limits. That usually means serving only tools that are valid for the caller’s current authority, hiding or suppressing unapproved tools, and ensuring the returned metadata does not become a shortcut around later authorization checks.
Discovery also needs to be operationally consistent. If the server changes tool availability, the published discovery view should change with it; stale or over-broad listings can mislead callers, increase accidental misuse, and make policy enforcement harder to reason about.
Where environments use registered or hosted MCP servers, governance over the server itself matters as much as the tool list. MCP Server Trust Assessment is a natural companion because it focuses on provenance, change control, and approval before connection.
Why Discovery Errors Matter Before Any Tool Runs
The danger in MCP tool discovery is that the exposure happens before execution. Once a caller learns that a powerful tool exists, it can shape later prompts, retries, or social engineering attempts around that capability, even if the final action is denied. Discovery therefore affects both confidentiality and downstream abuse potential.
In mature deployments, discovery should be treated as part of the trust boundary, not merely part of the user interface. If it is left open, the environment may still appear secure at execution time while quietly revealing enough structure to weaken that security in practice.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 and OWASP API Security Top 10 address the attack and risk surface, while NIST SP 800-53 Rev 5 and NIST Zero Trust (SP 800-207) set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | Tool discovery exposes agent authority boundaries and may reveal overbroad capabilities. |
| Recommendation — Constrain discovered tools to the caller's approved authority and context. | ||
| NIST SP 800-53 Rev 5 | IA-9 — Service Identification and Authentication | MCP tool discovery is tied to service-to-service access and authenticated capability exposure. |
| AC-3 — Access Enforcement | Discovery must enforce which tools a principal can see before use. | |
| Recommendation — Bind tool exposure to authenticated service identity and approved audience. Enforce visibility rules so unauthorized tools are not disclosed in discovery. | ||
| OWASP API Security Top 10 | API5 — Broken Function Level Authorization | Discoverable tools resemble callable functions whose exposure must match authorization. |
| Recommendation — Align discoverable functions with function-level authorization checks. | ||
| NIST Zero Trust (SP 800-207) | Zero Trust Architecture | Discovery should be context-aware and continuously authorized, not assumed by network reachability. |
| Recommendation — Evaluate each discovery response against the caller's current context and trust state. | ||
Related resources from NHI Mgmt Group
- How should security teams handle tool discovery for AI agents in MCP environments?
- What breaks when MCP tool discovery is not validated centrally?
- What breaks when MCP tool discovery is hardcoded into agent configuration?
- How should security teams implement a centralized MCP gateway so tool discovery stays manageable across many servers and agents?
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