Join our Newsletter — 33% off our NHI Course
Agentic AI & Autonomous Identity

Tool Manifest

← Back to Glossary
By NHI Mgmt Group Updated October 30, 2026 Domain: Agentic AI & Autonomous Identity

The structured record of an MCP server’s tools, including names, descriptions, schemas, and return types. In practice, it is the baseline an organisation should compare against to detect drift, hidden changes, or unauthorised additions before an agent is allowed to trust the server again.

What the tool manifest tells you

A tool manifest is the control point for what an MCP server says it can do. It gives clients a structured description of each tool, including its name, purpose, input schema, and return shape, so the agent or orchestrator can decide whether the server is offering the right capability, and whether that capability still matches what was approved.

Because the manifest is the declared interface, it is not just documentation. It is what downstream automation uses to build trust, select tools, and validate expectations before a call is made. If the manifest changes without governance, the server may still look legitimate while behaving differently.

Why drift in a tool manifest matters

The security issue is not only malicious change, but also unreviewed drift. A renamed tool, altered parameter schema, changed output contract, or newly added tool can create unsafe assumptions in clients that cache or trust the previous state. That is why manifest comparison is a practical baseline for change detection, especially in environments where tool use is delegated to agents.

In the tool-calling path, a manifest can be the difference between a constrained capability and an overbroad one. If the declared tool set expands, or if a description is rewritten to encourage different use, the surrounding trust model may quietly change even when the server address has not. AI Agent Tool and Plugin Security Guide covers the same tool trust problem from the agent side, including tool validation and permission scoping.

How organisations verify and compare it

Practitioners usually compare the live manifest against a known-good baseline captured during onboarding, review, or prior attestation. The main checks are straightforward: has a tool appeared, disappeared, or changed shape; do the descriptions still match approved use; and do the schemas still enforce the same inputs and outputs. Where the server is remote, the manifest should be treated as part of the server’s security posture, not as a static API contract that can be trusted indefinitely.

This is where environment inventory and server hardening intersect. A manifest is only useful if the organisation knows which MCP servers exist, which agents can reach them, and which upstream changes should trigger re-approval. Remote MCP Server Hardening Guide is the natural companion for the transport, authentication, isolation, and logging side of that control plane.

What a safe manifest lifecycle looks like

A sound lifecycle treats the manifest as an approval artefact that must be versioned, reviewed, and revalidated whenever tool definitions change. That includes new tools, removed tools, schema edits, return-type changes, and any shift in how the server presents capability boundaries. If the manifest is signed or otherwise integrity-protected, the goal is not only to know what changed, but to know that the change is the one you intended.

For organisations using agentic tooling, the manifest should also be aligned with permissioning. If the server advertises more than the agent is meant to use, the manifest becomes a policy mismatch, even if the tools are technically available. In practice, the manifest should support least privilege by making it obvious which tools are expected, which are sensitive, and which changes require human review before trust is restored.

Detection and response after a manifest change

When a manifest changes unexpectedly, the correct response is to assume the tool surface may have changed as well. That means pausing trust, rechecking the server against the baseline, reviewing recent tool calls, and looking for signs that an agent may have interacted with a modified or malicious tool definition. The same applies if tool names or schemas appear to be shadowed, duplicated, or substituted in a way that could redirect calls.

Detection should focus on diffing the manifest, checking whether newly advertised tools are actually authorised, and confirming whether any existing tool now points to a different behaviour than before. If the change is unexplained, the safer posture is to treat the server as untrusted until the manifest, the surrounding permissions, and the server’s own integrity have all been revalidated.

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 Non-Human Identity Top 10 address the attack and risk surface, while NIST SP 800-53 Rev 5 sets the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
OWASP Agentic AI Top 10ASI02 — Tool MisuseTool manifests define which tools an agent can invoke and trust.
Recommendation — Validate advertised tool surfaces before allowing agent tool use.
OWASP Non-Human Identity Top 10NHI-06 — Insecure Cloud Deployment ConfigurationsMCP servers expose tool surfaces that must be hardened and controlled.
Recommendation — Harden MCP server deployment and verify exposed tool interfaces.
NIST SP 800-53 Rev 5SI-7 — Software, Firmware, and Information IntegrityManifest drift is an integrity problem for the declared tool interface.
CM-8 — System Component InventoryA manifest is useful when tool assets and exposed capabilities are inventoried.
AC-6 — Least PrivilegeManifest-controlled tool exposure should constrain agent capability to need-to-use only.
Recommendation — Detect unauthorized manifest changes and restore trusted tool definitions. Maintain an inventory of MCP servers and their approved tools. Limit agent access to only the tools approved in the manifest.

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.

NHIMG Editorial Note
Reviewed and updated by the NHIMG editorial team on October 30, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org