PRAE / Product guide
Work through MCP.
See the evidence.
PRAE is a headless security assessment platform with a visual evidence workspace. Your agent handles the repeatable work. The browser helps you understand results and control access.
01 / Access
From subscription to first assessment.
- Create and verify your account.
Use your own work email. Registration does not create a charge. Each person has a separate identity.
- Register the application.
The individual plan is US$399 annually for one named person and one project. A project includes the same application's related repositories and development, staging and production environments.
- Review the plan and terms.
When paid sales open, review the annual total, renewal policy and applicable tax in Stripe Checkout. Only verified payment creates an entitlement. A checkout return page never grants access by itself.
- Install and approve the runner.
Download the signed package, installer, manifest, verification keys and your receipt. Verify the release and choose the files and environments it may examine.
- Connect your coding assistant.
Use the local runner's generated MCP configuration. Ask it to profile the system, run the assessment, explain the findings and open the schematic.
Owner preview
Install your signed runner.
If your workspace shows an active owner preview, sign in, select its project and open MCP connections. Download the runner package, installer, signed manifest, verification keys and license receipt into one folder. These files authorize internal evaluation for the named owner and project; they do not create a paid subscription.
Use Node 24.22.2 or a newer Node 24 release. From that folder, install into a new directory:
node install-preview.mjs --archive nullventures-security-platform-0.3.0-preview.5.tgz --manifest null-security-release.json --trust null-security-trust.json --license null-security-license.json --directory ./prae-installationVerify the signing-key fingerprint through the approved owner channel before trusting the downloaded keys. Run the generated prae-installation/null-security.mjs launcher with doctor, then draft your scope using the exact project ID shown in your workspace. The included installation guide covers scope approval, your local MCP connection and backup. Receipts expire within seven days; renew them from the workspace.
02 / Connections
Two MCP deployments. One project boundary.
Local or private MCP runner
The primary assessment interface. It runs inside the development computer, CI environment or private server that you authorize. It reads only approved files. It does not execute arbitrary repository code.
Use it while building in a coding assistant or against a checked-out version of an already deployed application. A repository assessment does not prove what is currently running in production; deployment and runtime evidence must be collected separately.
Its 27 tools cover profiling, controls, assessments, findings, remediation planning, separate verification, evidence integrity, approved workflow tests, Flow Map synchronization, requirements, change comparison, Vigilance, report export, the local artifact library, diagnostics and explicit sharing.
Hosted MCP review
Add the server URL below to a compatible MCP client. Sign in through the browser, review the client and select one authorized project. The client receives its own OAuth token; you do not share a password.
https://nullventures.co/api/platform/mcp
The hosted interface reads an approved snapshot shared by your runner. Ten tools expose connection health, a private map link, Flow summary, data journeys, requirements, changes and the shared artifact library. List, inspect, open or compare records through the same project access used in the browser. It cannot run a scan on a computer it cannot reach.
For clients without supported browser authorization, create a named, project-bound agent credential in MCP connections. Enter it in the client's secret configuration as a Bearer token. Runner credentials have a different role and cannot be used as agent credentials.
Useful things to ask your agent
“Profile this approved project. What is the stack, and which protections need evidence?”
“Run the assessment. Explain the highest-priority confirmed gaps separately from untested controls.”
“Open the private infrastructure map and trace this dataset through its journey.”
“Prepare a correction for this finding. Tell me what separate test would verify it.”
“Recheck the changed requirements. Do not close the finding without the required human decision.”
In a browser-only building platform, use hosted review unless that platform can run or connect to your private runner. A live application on Vercel, Netlify or another server still needs explicit deployment bindings and approved tests. Connecting MCP does not grant access to a cloud account.
03 / Product surfaces
What happens where.
| Task | Local / private MCP | Hosted MCP | Browser workspace |
|---|---|---|---|
| Identify stack and assess approved code | Primary workflow | Reads shared results | Shows shared artifacts |
| Refresh data map and recheck requirements | Available within approved scope | Reads latest shared snapshot | Shows capture time and coverage |
| Run a reviewed runtime test | Approved test/staging plans only | Unavailable | Inspect resulting evidence |
| Explain findings and prepare corrections | Available; no automatic deployment | Requirements and next actions | Visual context and evidence |
| Open a schematic or data journey | Returns a private local or hosted link | Returns a sign-in-protected link | Full graphical view |
| Approve new scope or close findings | Human/operator action outside tools | Unavailable | No automatic approval |
| Subscribe and manage billing | Links to account | No payment authority | Verified account and Stripe portal |
| Manage people, projects and credentials | No administration authority | No administration authority | Role-checked workspace controls |
| Monitor local evidence and recovery | Health tool; operator backup commands | Shared-evidence freshness | Map coverage; owner operations view |
04 / Governance
A subscription is an assigned right.
Individual
One verified person, one registered application. Hosted requests recheck the entitlement and project binding. Creating another account or changing a project label does not transfer the subscription. Related environments belong to the same project; unrelated products do not.
Team
US$1,999 annually for five named people and five registered projects, including the billing owner. Owners and administrators invite verified recipients and assign projects. Members can connect runners for their assignments. Viewers can read shared evidence. Only the owner can grant an administrator role.
Credentials
Manual credentials are shown once and stored as hashes. Each has a person, project, purpose, name and expiry. At most two active credentials of each purpose are allowed per person and project. Rate limits and issuance limits apply independently of an IP address.
Revocation
Removing a member or changing their permissions revokes their manual keys. OAuth requests recheck membership, project assignment and active access; account owners can revoke their application connections. Offline receipts are limited to seven days or the remaining paid period.
05 / Evidence
A missing control should be unmistakable.
The map connects components and data journeys to their requirements. A confirmed gap is red. An unknown or untested protection is amber and explicitly labeled. Verified results use the same restrained black-and-white treatment as the rest of the workspace.
For example, a route without observed rate-limiting protection can produce a scoped requirement, a Cloudflare rule draft or application-level correction plan, and a separate verification step. A drafted rule is not a deployed control. A source-code pattern is not runtime evidence.
The engine preserves scope, capture time, source revision, deployment binding, configuration evidence and runtime observations. Stale evidence stays visibly stale. It does not infer sensitive-data lineage from a generic service dependency.
Explore the fictional Flow Map ↗ Explore the fictional system map ↗Examples are evaluated by the real renderer and engine over clearly labeled fictional inputs. They are not an assessment of your application.
Project evidence
One library for the work your agent produces.
Open Evidence in your workspace to browse schematics, assessments, changes, correction plans, verification and diagnostics. Filter by type or environment, inspect capture and source information, compare two records, or download HTML, JSON or Markdown. Local MCP can open retained artifacts on your computer. Hosted MCP reads the records you explicitly shared.
Hosted records expire 24 hours after capture. Local history remains available in the runner. Older map-only shares remain in Overview until refreshed with the current named runner.
Connect the evidence library →06 / Operations
Plan for interruption.
Keep the runner's evidence store on an encrypted volume and retain private backups and the fingerprint key. Verify a restored workspace before replacing the working one. Preserve the previous installation during upgrades and verify signed manifests before switching your MCP command.
The hosted service uses server-checked sessions, restricted database roles, audited access changes, bounded reconciliation jobs and encrypted backup support. Release readiness includes deployed tests, external monitoring and a restore exercise. The owner operations view reports the recorded state; an unperformed check remains open.
Customer-specific cloud discovery and internal network testing need explicit permissions and supported adapters. Runtime interception/enforcement is a separate milestone. PRAE's current assessment and simulation tools are not an enforcement gateway.