Resolute Security MCP server

Current security, compliance, and insurance posture scores for your organization. MEASURED: each score is computed by the same pure scoring engine the dashboard uses (lib/scoring), from live scan results, finding counts, framework responses, MFA enrollment, and DMARC policy. A null score means there isn't enough data yet — never assume zero. No arguments; the org is taken from your API key.

How your posture scores have moved over time. Returns the latest-vs-prior delta plus a daily history window. MEASURED from stored daily snapshots (score_snapshots); if snapshots are sparse the delta uses the closest earlier snapshot. `daysAgo` sets the delta baseline (default 7); `historyDays` sets the history window (default 60).

Your ranked 'what to do next' list — the same engine that powers the dashboard's next-actions and the AI coach. MEASURED: derived from open findings + SLA state, failing insurance checklist items, in-progress frameworks, integration errors/gaps, overdue offboarding, and vendor signals. No arguments.

Your cyber-insurance underwriting readiness: the twelve underwriter-checklist controls plus the impact-weighted readiness report (score 0–100, hedged band summary, gaps ranked by pointsAtStake, per-coverage roll-up, and the supplemental-application Q&A). Each control row separates what is MEASURED from what is ATTESTED: `auto` {state, reason, sources (fact keys), lastCollectedAt} is derived from live integration facts; `manualOverride` is the operator's attestation (null when none); `effective` is what the report scores (override wins). HONESTY, built in: the auto-fill deliberately EXCLUDES facts recorded via MCP attestation (source 'mcp') — an agent's own profile_fact_set writes can NEVER inflate what this tool reads back. A 'no-data' control counts as NOT met (you can't attest to what you can't evidence). The score is an internal readiness signal, not a carrier rating, and never promises a premium or a bind. READ-ONLY; internal source references are never returned. No arguments.

List your organization's findings with optional filters and cursor pagination. Filters match the findings board exactly: `status` (comma-separated: open,in_progress,resolved,risk_accepted,wont_fix), `severity` is a FLOOR (e.g. 'high' returns high+critical), plus `kind`, `source`, and `q` (substring on title/description). Pass `cursor` from a prior page's nextCursor to continue. Org-scoped; secrets and free-form metadata are never returned.

Fetch a single finding by id, scoped to your organization. Returns not_found (404) when the id doesn't exist OR belongs to another org — the two are indistinguishable on purpose. Secrets and free-form metadata are never returned.

FILES an observed issue, risk, or gap as a tracked FINDING in the platform's unified findings list (the same list scans + integrations feed). Use it to record something you discovered while connected to OTHER sources — e.g. a misconfiguration seen via an M365/SharePoint MCP, or a risk noticed on a web page — so it becomes actionable and audit-trailed. HONESTY: record only what you actually observed; never invent a finding or inflate severity. It is stored with source "mcp" so it stays clearly distinguishable from a measured scan, and can be linked to compliance controls via controlRefs. REQUIRES THE "write" SCOPE (read-only connections are rejected); the org + author come from your credential, never an argument. Idempotent: re-filing with the same `ref` UPDATES the existing finding (and re-opens it if it had been resolved) instead of duplicating. `severity` is info|low|med|high|critical.

Move 1–200 findings to a new lifecycle status (open | in_progress | resolved | risk_accepted | wont_fix) — the same transition rules as the findings board: terminal statuses stamp resolvedAt/resolvedBy/resolvedNote, re-activating clears them, and a finding already in the requested status is a no-op reported as applied. STRICTER THAN THE UI, on purpose: `note` is REQUIRED and must be non-empty for resolved, risk_accepted, and wont_fix — every terminal transition an agent makes carries an on-the-record justification. Ids are org-scoped in one query; an unknown or foreign id is reported not_found in its per-id outcome (indistinguishable on purpose). One audit row is written PER changed finding (finding.status_changed, previous→new, via:'mcp'). REQUIRES THE "write" SCOPE on a user-authorized connection; demo sandbox blocked; metered by the org-day MCP write quota. Returns {applied, notFound, invalid, outcomes[]}.

CMMC readiness scorecard for a level (L1 | L2 | L3). MEASURED and EXACT: this is a count/ratio of control responses, NOT a weighted SPRS score — readiness is the share of controls answered 'yes' (met), and openGaps are controls answered 'no' or 'partial'. Controls with no answer count as unanswered, never as a gap. Reuses lib/cmmc/scorecard.computeScorecard over a read-only loader (no assessment is created). If you haven't started this level, every control is unanswered.

Your raw CMMC control responses for a level (L1 | L2 | L3), plus per-family progress. MEASURED: returns each answered control's status (yes|partial|no) and notes, read-only — it never creates or modifies an assessment. Unanswered controls are simply absent. Use framework_get_scorecard for the rolled-up readiness number.

Per-control evidence worklist for a framework — what proof is still needed. MEASURED from current data and READ-ONLY: it does a write-free read of your existing assessment and NEVER creates one. If you haven't started a framework, every control is reported as 'unanswered' and the summary says so. HONESTY: a control counts as 'present' ONLY when it's answered 'yes' AND backed by at least one FRESH evidence row (within the 1-year freshness window); 'yes' with no proof is 'missing', and 'yes' with only old proof is 'aging'/'stale' — never implied satisfied without current evidence. Anything not answered 'yes' is an assessment gap ('unanswered'), surfaced separately from evidence gaps. `framework` is one of cmmc|soc2|iso27001|hipaa|pci|gdpr|csf; `level` (L1|L2|L3) applies to CMMC only and is ignored by the others. No file blobs or secrets are ever returned.

One-glance overview of every compliance framework: whether your org has started it, an honest readiness percent, and a quick evidence-gap count. MEASURED and READ-ONLY: each framework is resolved with a write-free assessment read (no assessment is ever created) and readiness reuses the SAME per-framework readiness engine the dashboard uses. An unstarted framework returns started=false and readinessPercent=null — never a fabricated number. gapCount counts controls answered 'yes' without fresh evidence PLUS controls not yet answered 'yes' (it matches compliance_get_evidence_gaps exactly). For CMMC, the readiness reflects the level you've actually started (preferring L2). No arguments; the org comes from your credential.

ATTACHES one piece of proof (a link, a note, and — for CMMC only — a file) to a single control in a framework's evidence library. HONESTY: this ONLY adds evidence. It does NOT change the control's answer, readiness, status, or score — deciding whether a control is "met" stays a human decision. It cannot edit or delete anything. REQUIRES THE "write" SCOPE: connect with write access allowed (read-only API keys and read-only connections are rejected). The evidence is recorded as added by the authorizing user, and the org comes from your credential (never an argument). `framework` is cmmc|soc2|iso27001|hipaa|pci|gdpr|csf. `controlId` is the control id (for CSF the subcategoryId, for SOC 2 the criterionId). `level` (L1|L2|L3) applies to CMMC only (default L2). `label` is a short description (required); `url` (http(s) only) and `note` are optional. `file` (name+contentType+base64) is CMMC ONLY and is size-/type-limited; file bytes are never returned. Idempotent: re-submitting the same (control, label) within 60s returns the existing evidence with deduped:true instead of creating a duplicate. Submitting evidence for a control you have not started will create that framework's assessment (then attach the evidence).

Render an assessor-ready CMMC artifact for a level (L1|L2|L3) and return it INLINE as text: `artifact` is 'ssp' (System Security Plan, markdown), 'poam' (Plan of Action & Milestones, CSV), or 'package' (SSP + POA&M combined markdown — the default; what you hand a C3PAO). MEASURED and READ-ONLY: it renders your real control responses, evidence labels/links (file BLOBS are never embedded), CUI scope assets, and the POA&M columns are folded from your actual findings-board triage (due dates, owners, tracking status). The L2 SSP includes the estimated SPRS score computed per the DoD methodology — verify before submitting to SPRS. Resolution is write-free: if you haven't started the level, it returns started:false and an honest artifact showing every control unanswered — nothing is created and nothing is invented. Returns {artifact, level, started, contentType, filename, content}. The export is audit-logged.

Record an answer (yes | partial | no) for ONE control in a compliance framework's assessment. HONESTY CAP, stated plainly: an agent-submitted 'yes' is STORED AS 'partial' with an "[attested:mcp]" note marker and capped:true in the result — an AI can record gaps and progress but can NEVER single-handedly mark a control met; a human confirms 'yes' in the app. 'partial' and 'no' are stored unchanged (they can only lower or hold readiness). OVERWRITE GUARD: by default an existing answer BLOCKS the write (only_if_unanswered defaults true); pass only_if_unanswered:false explicitly to replace it — the cap still applies. `framework` is cmmc|soc2|iso27001|hipaa|pci|gdpr|csf; `controlId` is validated against that framework's real catalog (CSF: subcategoryId; SOC 2: criterionId); `level` (L1|L2|L3) is CMMC-only, default L2. Answering a framework you haven't started creates its assessment (assessmentCreated:true). REQUIRES THE "write" SCOPE on a user-authorized connection; demo sandbox blocked; metered by the org-day MCP write quota. The answer is stored as updated by the authorizing user, and the write is audited via:'mcp' with requested vs stored status on record. Returns {requestedStatus, storedStatus, capped, previousStatus, assessmentCreated}.

Run the SAME 'Pre-fill from integrations' engine as the in-app button: for every CMMC control your connected integrations' MEASURED facts can support, suggest (dry-run) or record (apply:true) an answer, attaching the driving fact as evidence. DRY-RUN BY DEFAULT: apply defaults to false and the dry-run writes NOTHING (read-only resolution — it will not even create the assessment); it returns the engine's suggestions verbatim ({controlId, status, factKey, rationale, attested}) plus wouldApply/skippedExisting counts. HONESTY, enforced by the engine and this tool: statuses come FROM THE ENGINE only (you may filter WHICH controls via controlIds, never choose a status); AI-attested facts are already capped to 'partial' by the engine (attested:true explains why); an apply SKIPS every control that already has an answer — it never overwrites a human, and there is no override parameter. apply:true REQUIRES THE "write" SCOPE on a user-authorized connection; demo sandbox blocked; metered by the org-day MCP write quota; ONE audit row (cmmc.prefilled_from_integrations, via:'mcp') with the counts. Applied answers are noted "[auto:integration] <rationale>". CMMC-only for now. `level` is L1|L2|L3.

Cross-integration overview for your organization: connection coverage by category, a health roll-up (healthy/stale/error) over connected sync providers, the signal each connection earns you, and a ranked 'connect next' list. MEASURED: health verdicts reuse the same freshness logic the dashboard uses. SECRETS REDACTED: encrypted credentials, tokens, and secrets are never returned. No arguments.

Trigger a manual data sync for ONE connected integration, by `provider` (the org's integration for that provider is resolved automatically — preferring a healthy connection, then a stuck one to retry) or by `integration_id`. Pass exactly one of the two. OWNER-ONLY — the org's role model has two roles, owner and member, and this is the first owner-gated tool: a manual sync consumes shared worker capacity and calls the provider's API, so it stays with owners (members get a 403). Also requires the "write" scope and a user-authorized (OAuth) connection; blocked for the demo sandbox; metered by the org-day MCP write quota. The trigger itself reuses the app's Sync-now path verbatim: only active|error integrations sync (pending/disconnected must finish reconnecting first), a 5-minute per-integration cool-down applies ({ok:false, reason:'cooldown', retryAfterMs} — not an error), and a time-bucketed job id dedups back-to-back triggers. HONESTY: this ENQUEUES a background job and returns immediately — there is no job-status tool; poll integrations_get_overview and watch the connection's freshness/health. Audited via:'mcp'.

Your organization's user-access-review campaigns. Without `campaign_id`: an org-wide summary (total/active/overdue campaigns, pending items, last completion) plus every campaign with its live review progress; `status` (draft|active|completed) optionally filters that list. With `campaign_id`: one campaign's full detail — items ordered pending-first, each with subject email/name, source (m365|team), decision (pending|keep|revoke|flag), decision notes, reviewer, and review time. MEASURED with one caveat stated plainly: each item's `context` (isPrivilegedAdmin, isMfaRegistered, mfaMethodCount, accountEnabled, lastSignInAt, forwardingEnabled) is a SNAPSHOT taken when the campaign was created — it is NOT live M365 state and does not update as the tenant changes. READ-ONLY. Returns not_found (404) when the campaign id doesn't exist OR belongs to another org — indistinguishable on purpose.

Create AND launch a user-access-review campaign: subjects are snapshotted from the chosen sources — 'm365' (the connected Microsoft 365 directory, with per-user risk context) and/or 'team' (this platform's own member roster); default is both — and each becomes a pending review item. NOT IDEMPOTENT: every call creates a NEW campaign, so call access_review_get first and reuse an active campaign instead of stacking duplicates. itemCount is reported honestly — 0 means the chosen sources yielded no reviewable subjects (e.g. no active M365 integration), not a failure. The subject context is snapshotted at THIS moment and does not update afterwards. Requires the "write" scope on a user-authorized connection AND a Pro subscription; demo sandbox blocked; metered by the org-day MCP write quota. `due_date` is "YYYY-MM-DD" (optional); `scope_notes` is free-form reviewer guidance (optional). Audited via:'mcp'. Record decisions with access_review_record_decision.

Record the review decision for ONE campaign item: 'keep' (access is appropriate), 'revoke' (access should be removed), or 'flag' (needs human follow-up). An agent cannot set an item back to 'pending' — that would erase a reviewer's work. `notes` (≤2000 chars) is REQUIRED and must be non-empty for 'revoke' and 'flag': the justification is what makes the review defensible to an auditor. LOAD-BEARING HONESTY: 'revoke' records the review VERDICT only — it does NOT deprovision, disable, or change the account anywhere; actual removal happens in the IdP (e.g. M365 admin center) and is verified by the next directory sync. Re-recording a decision overwrites the previous one (latest reviewer wins, stamped with your user). The campaign's findings are re-projected after each decision, and the fresh campaign progress is returned so you can see how much review remains. Unknown/foreign item id → not_found. Requires the "write" scope on a user-authorized connection; demo sandbox blocked; metered by the org-day MCP write quota. Audited via:'mcp'.

Every domain your organization monitors, with the score from its latest scan. MEASURED: latestScore is the overall score of the most recent scan (null when a domain has never completed one). Org-scoped; no arguments.

Aggregate DMARC posture for ONE policy domain over a recent window. MEASURED and READ-ONLY: rolls up stored DMARC aggregate reports into reportCount, the report date range, the most-recently published policy (p/sp/pct), message stats (total, aligned, alignmentPct), a disposition tally for failing mail, and the top 5 source IPs by failing-message count. `policyDomain` is required; `daysBack` (default 30, max 365) bounds the window. Message-level PII (header-from / envelope addresses, raw XML) is never returned. Returns a well-formed empty summary when there are no reports for that domain in-window.

Fetch a single monitored domain by id, scoped to your organization, with its latest-scan SUMMARY: overall score, scan status, completion time, plus isActive and alertsEnabled. MEASURED and READ-ONLY. The full per-check scan-results blob is omitted — only the summary is returned. Returns not_found (404) when the id doesn't exist OR belongs to another org.

Enqueue an immediate security rescan for ONE monitored domain (by its UUID from monitoring_list_domains). Same rules as the app + API rescan path, exactly: org-scoped (unknown/foreign id → not_found), a shared per-domain 10-minute cool-down ({enqueued:false, reason:'cooldown', retryAfterSeconds} — not an error), and the org-day manual-rescan quota (20/day; typed quota_exceeded with retryAfterSeconds when exhausted) layered on top of the MCP write quota. Requires the "write" scope on a user-authorized connection; demo sandbox blocked. HONESTY: this ENQUEUES the scan and returns {enqueued:true, jobId, domain} — there is NO job-status tool; poll monitoring_domain_get for this domain and watch its latest-scan completion time advance (typically 1–3 minutes). Audited via:'mcp'.

Draft candidate answers for security-questionnaire questions from your org's own material — review before sending. For each question (1–200 per call) it returns: `bestMatch`, the closest answer from your saved vendor-answer library (same Jaccard matcher and 0.3 threshold the questionnaire upload uses; below threshold → null — a fabricated answer is never returned), with the matched library question, saved answer, evidenceRef, and a MEASURED similarity score (token overlap, not semantic understanding); and `factSuggestions`, templated answer sentences derived from LIVE collected posture facts (e.g. MFA enrollment %, DMARC coverage), each naming its factKey, source, and keyword-match confidence. A fact suggestion only appears when the org has actually collected that fact. READ-ONLY: persists nothing — these are drafts for a human to review, edit, and send.

Grant a prospect access to your trust page's gated documents (SOC 2 report, ISO cert, pen test, …): issues a unique 30-day access link for their email and (by default) emails it to them — the same flow as the in-app 'Invite' button, including its per-org daily invite quota. The RAW LINK IS RETURNED EXACTLY ONCE in this result (shareUrl): the token is stored hashed and cannot be recovered later — only revoked and reissued. Grants are revocable any time from the trust admin page (Posture → Trust page). If the email send fails the grant still stands (emailSent:false + emailError) and you can deliver shareUrl yourself. Set send_email:false to skip the email and just get the link. Requires an existing trust page and THE "write" SCOPE on a user-authorized connection; demo sandbox blocked; metered by the org-day MCP write quota. Audited via:'mcp'.

Every third-party vendor your organization tracks, best security score first. MEASURED: scores (0..100) are cached from the latest runScan() pass against each vendor's primary domain; a null score means 'not scanned yet', not zero. Org-scoped; no arguments.

Fetch a single tracked third-party vendor by id, scoped to your organization, with its sub-scores (overall/email/web/dns) and its attestation rollup (status: none|fresh|stale, last-attested date, daysStale past the annual mark). MEASURED: scores are cached from the latest runScan() pass — a null score means 'not scanned yet', not zero. READ-ONLY. The full scan-results blob, free-form notes, and raw questionnaire responses are never returned. Returns not_found (404) when the id doesn't exist OR belongs to another org.

Per-vendor attestation rollup for EVERY tracked vendor: {vendorId, vendorName, primaryDomain, attestationStatus, lastAttestedAt, daysStale}. MEASURED and READ-ONLY from the latest completed attestation per vendor. attestationStatus is 'none' (no attestation on file), 'fresh' (within 1 year), or 'stale' (older than 1 year — auditors expect a refresh); daysStale counts days past the 1-year mark (0 unless stale). Raw questionnaire answers are never returned. Org-scoped; no arguments.

List your organization's written policies (governance documents) with optional status filter and cursor pagination. MEASURED and READ-ONLY: a direct, org-scoped read of the policies table. Returns metadata only — id, title, one-line description, status (draft|active|archived), version, frameworks satisfied, owner email, template slug, review-interval days, and last-reviewed/created/updated timestamps. The full policy BODY is omitted here (it can be large) — use policies_get for one policy's body. Pass `cursor` from a prior page's nextCursor to continue.

Fetch a single policy by id, scoped to your organization, INCLUDING its full markdown body (bodyMd). MEASURED and READ-ONLY. Returns not_found (404) when the id doesn't exist OR belongs to another org — the two are indistinguishable on purpose (anti-enumeration). No secrets are returned.

List the members of your organization: userId, email, name, org role, joinedAt, and mfaEnrolledAt (null = MFA not enrolled). MEASURED and READ-ONLY from memberships joined to users. SECURITY: password hashes, the MFA secret, password-lockout/attempt counters, email-deliverability forensics, and notification preferences are NEVER returned — only identity, role, join time, and the MFA-enrollment signal. Org-scoped; no arguments.

Your organization's profile: name, slug, subscription status (a coarse plan signal), onboarding completion + goals, created date, and — when filled in — industry, size band, primary jurisdiction, privacy obligations, and profile notes. MEASURED and READ-ONLY (organizations + company_profile). SECURITY: Stripe identifiers, Slack/generic webhook URLs and their signing secret, the inbound DMARC mailbox token, and the inline logo blob are NEVER returned. No arguments; the org comes from your credential.

Auto-collected posture facts about your organization, keyed by a dotted fact key (e.g. idp.mfa_enrolled_pct, endpoints.encrypted_pct): {factKey, value, source, collectedAt}. MEASURED and READ-ONLY from profile_facts — `value` is the latest typed value an integration (or manual entry) reported. The internal sourceRef is omitted. Org-scoped; no arguments.

Your organization's append-only audit trail, newest first, cursor-paginated. MEASURED and READ-ONLY: each entry is {id, action, target, userEmail (resolved from userId; null for system actions or a deleted user), createdAt, metadata}. The free-form metadata is SECRET-REDACTED and capped to a handful of keys, so verbose or sensitive payloads can't leak. Optional filters: `action` (exact match, e.g. 'user.signed_in') and `daysBack` (max 365). Pass `cursor` from a prior page's nextCursor to continue.

RECORDS one organization posture fact (e.g. "MFA enforced for all users") into the org profile. It is stored as an ATTESTATION (source "mcp"): it surfaces in the profile and can SUGGEST "partial" readiness for the CMMC controls that fact maps to (a human still confirms), but it NEVER auto-confirms a control as "met", never satisfies a live-feed freshness gate, and is excluded from insurance auto-confirm. Use it to capture a fact you actually verified from another connected source (M365, an admin console, a document). REQUIRES THE "write" SCOPE; org comes from your credential. Idempotent on factKey (latest write wins). Prefer keys already in use — call profile_facts_list first to see them (e.g. "idp.mfa_enrolled_pct"); ad-hoc keys are stored but may not surface in the UI/evidence flow.