MCP Capabilities

The Tool Catalog is the exact reference — every argument and result shape. This page is the map: what an AI client connected to AutoCISO can actually answer, grouped by the question you’d ask.

All 23 tools are organisation-scoped and PII-masked. 22 are read-only; one writes.

1. Access intelligence

“Who has access they shouldn’t?” — the largest group, and the core of the product.

Nine tools cover stale and orphaned accounts, privilege concentration, privileged-access changes over the last 30 days, employees in high-risk joiner/mover/leaver states, anomalous access patterns, offboarding violations, per-department risk scoring, security-debt trend over time, and access-data normalisation health.

“Which accounts are stale, and which departments do they cluster in?”

Tools: access_stale_accounts, access_concentration, access_privileged_delta, access_high_risk_lifecycle, access_anomaly_discovery, access_offboarding_violations, access_department_risk, access_security_debt_trend, access_normalization_health

2. Access reviews

“Where does my review campaign stand?”

Campaign status and completion progress for user access reviews (UAR).

“Is the Q3 access review on track, and what’s outstanding?”

Tools: uar_status, access_reviews_list

3. ISO 27001 posture

“What are my gaps, and what’s my Statement of Applicability status?”

Gap analysis, SoA summary, and the full control list with per-control gap status. All three accept an optional projectId and fall back to your first active ISO project.

“Summarise my open ISO 27001 gaps by Annex A theme.”

Tools: iso_gap_status, iso_soa_summary, iso_controls_list

4. Risk register

“What’s on the register — and file this as a draft.”

Read the register, fetch one risk in full, and create a new risk as a draft. This is the only group with a write capability.

“List open high-severity risks, then draft a new one for the finding we just discussed.”

Tools: risk_register_list, risk_get, risk_create_draft

5. Supplier and third-party risk

“Which suppliers are unassessed?”

The supplier list and assessment-status rollup.

“Which tier-1 suppliers still have an incomplete assessment?”

Tools: supplier_list, supplier_assessment_status

6. Audit and evidence

“Show me the trail.”

The evidence log and a searchable audit trail filterable by exact actor ID.

“What did this actor change last month?”

Tools: evidence_log, audit_trail_search

Mind the result shapes here: audit_trail_search paginates and returns { items, total, limit, offset }, while evidence_log returns at most 50 entries with no continuation token — a silent cap. The Tool Catalog flags every capped tool.

7. Executive posture

“Give me the board-level summary.”

Maturity scores per domain and a single-call posture summary — overall score, stale accounts, open risks, open ISO gaps, and the top concern. posture_summary returns useful data even for a fresh org, which makes it the best first call.

“Give me a one-paragraph security posture summary for the board deck.”

Tools: maturity_scores, posture_summary

What MCP will not do

• No cross-tenant access. No tool takes an org identifier, and there is no header to set one. Your organisation is bound into the token and resolved server-side. This is enforced structurally: the server refuses to start if any tool declares an orgId-shaped input field.
• No URL fetching. No tool accepts a URL, endpoint, or callback argument — also enforced as a startup check, which closes off server-side request forgery.
• No raw PII. Results are masked before they leave the server: emails become a***@domain.com and personal names are redacted.
• No bulk export. Tools answer questions; they are not a data-export channel. Several cap their results.

Next steps

• MCP Quickstart — first call in five minutes
• MCP Tool Catalog — exact signatures
• Connect Claude Desktop