Your compliance data in your AI tools

The Model Context Protocol server lets clients like Claude and Cursor work directly against your tenant's compliance graph, securely scoped to your organization, with read tools that answer and action tools that only ever propose.

Back to AI

ClaudeCursorAny MCP client

compliance-mcporg-scoped

> list_failing_controls

2 controls need evidence:
  • CC6.1  Encryption at rest
  • CC7.2  Log review cadence

scope: org:your-tenant
(never crosses tenants)

What it can do

Tools your AI client can call

Read your compliance graph

List frameworks, controls, requirements, and the evidence behind them, straight from your AI client, without a context-switch into the app.

Surface failing controls

Ask which controls need attention and what evidence is missing, and get an answer scoped to your tenant without leaving your editor.

Draft from your data

Generate questionnaire answers, summaries, and remediation notes grounded in your own policies and test results, as drafts your team reviews.

Tenant-scoped, always

Every tool call is scoped to your organization by an org-bound token. The server cannot reach another tenant's data, by construction.

Tool reference

Read tools answer, action tools propose

The split is deliberate. Reading your posture is safe to do directly; anything that could change something is staged for a human.

Read tools

direct

list_controls
All controls and their current status across active frameworks.
control_status
Framework-level readiness rollups: pass/fail/warning counts and percent ready.
failing_tests
Tests currently failing, with the controls they affect.
get_evidence
The evidence backing a control or test, with source and timestamp.
query_compliance
Natural-language query over your own graph, grounded with citations.
list_risks
Your risk register: likelihood, impact, inherent/residual scores, owner.
list_vendors
Your third-party risk register with the latest SOC 2 / risk assessment per vendor.
list_policies
Your policy library: version, active flag, and review cadence.
list_findings
Internal findings open for triage, filterable by status or severity.
list_access_reviews
Access-review campaigns, or the reviewer decisions inside one.
list_questionnaires
Uploaded security questionnaires and answer-approval progress.
list_trust_center_docs
Your trust-center document library across all visibilities.
list_integrations
Connected integrations with asset compliance and evidence provenance.
list_training_assignments
Employee security-awareness training assignments and completion.
list_ai_systems
Your AI system inventory: vendor, purpose, risk tier, human oversight.
list_exceptions
Your exception register: time-boxed waivers, with status and expiry.
list_review_schedules
Your recurring review-cycle queue across all reviewable subject types.

Action tools

propose-only

draft_remediation
Propose a remediation for a control or failing test, returned as a draft, never applied.
open_finding
Stage a finding for a human to confirm before it enters the findings inbox.
draft_policy
Propose a policy document for a topic, returned as a draft, never saved.
propose_risk_treatment
Propose a treatment plan for an existing risk, never applied to the risk record.
draft_questionnaire_answer
Propose an answer to a questionnaire question, grounded in your data, never saved as final.
propose_access_review_decision
Propose an approve/revoke recommendation for one access under review.
request_exception
Request a time-boxed exception; creates a DRAFT row awaiting human approval.

Action tools never mutate a connected system or send anything externally. They stage a draft or a finding; a human confirms.

Connect in minutes

Three steps to wire it up

Generate a scoped token

Create an org-scoped credential from your workspace settings. It grants access to your tenant only, with scopes you choose, and can be revoked at any time.

Add the MCP server

Point your MCP client at the server URL and paste the token into your client's config. Claude, Cursor, and any standards-compliant MCP client work the same way.

Ask in plain language

Query posture, list failing controls, or draft answers. Read tools return data directly; action tools return proposals your team still reviews before anything ships.

Security model

Scoped to your org by design

Org-scoped tokens

Each credential is bound to one organization and a set of scopes. There is no token that can read across tenants, and tokens are revocable and have a last-used timestamp.

Same isolation as the app

MCP tool calls go through the exact tenant-isolation layer the web app uses. The server is not a back door around your data boundary; it inherits it.

Read-biased, approval-gated

Most tools are read-only. The few action tools are propose-only: they stage drafts and findings for a human to confirm, and never mutate a connected system on their own.

FAQ

The MCP server, answered

The Model Context Protocol is an open standard that lets AI clients call external tools. By shipping an MCP server, we let your existing AI tools (Claude, Cursor, others) work directly against your compliance data, instead of you copying context back and forth. Because we are AI-native, this is a natural fit rather than a bolt-on.

No. The bulk of the tools are read-only. The action tools (draft_remediation, open_finding, draft_policy, propose_risk_treatment, draft_questionnaire_answer, propose_access_review_decision, request_exception) are propose-only: they stage a draft, proposal, or a single internal row for a human to review and confirm. No MCP tool mutates a connected system or sends anything externally without an explicit human gate.

Every tool call carries your org-scoped token and runs through the same isolation layer as the web application. A credential for one organization can never read another's data; isolation is enforced at the data layer, not just hidden in the UI.

Revoke it from workspace settings; it stops working immediately. Tokens are scoped, so a leaked token only ever had access to one org's data and only the scopes you granted it. Last-used timestamps help you spot a token that is being used unexpectedly.

Any standards-compliant MCP client. Claude and Cursor are the common ones, but the server speaks the protocol, so anything that implements MCP can connect with the same token-and-URL setup.

No. The in-app AI assistant works without it. The MCP server is for teams who would rather work from their own AI tools; it brings the same grounded, review-gated workflow into your editor.

Put compliance in your AI workflow

View a demo and we'll show the MCP server connected to a live tenant.

View a demo Explore the AI assistant