Security Preflight · v1.2

Check the MCP integration.
Before you connect it.

Inspect the manifest, tool schemas, authority policy, and sample text you supply. Get a signed assessment that makes declared risks and gaps visible before an integration changes.

Start an assessment Try the free change check

$1 USDC per paid assessment. An eligible introductory quote may be $0.01. The fresh quote governs the price; the buyer authorizes every purchase.

Static artifact assessment. It does not fetch, execute, or certify the deployed runtime.

What the assessment covers

Make declared authority inspectable.

Manifest and schemas

Endpoint and authentication declarations, tool input schemas, and whether inputs are bounded by the declared schema.

Tool policy

High-impact actions, approval requirements, and conflicts between allowed, denied, and approval-required tools.

Sample text

Static injection indicators in the supplied text. Coverage is limited to the artifacts you provide.

What you receive

A result you can inspect and retain.

A verdict and individual findings.
Digests binding the assessment to your manifest, policy, and samples.
A signed, input-redacted assessment receipt.
A delivery receipt on a successful paid HTTP response.

Raw supplied artifacts are not stored in the assessment record. A receipt proves its signed contents; buyer acceptance and real-world usefulness are separate judgments.

Assessment structure

{
  "verdict": "pass | review | fail",
  "evidence": {"checks": ["individual findings"]},
  "receipt": {
    "receipt_id": "returned assessment ID",
    "evidence_sha256": "bound evidence digest",
    "expires_at": "assessment expiry"
  },
  "signature_b64": "returned signature"
}

Shortened response shape, not an executed assessment. Read an unsigned filtered public receipt summary at /security-preflight/receipts/{receipt_id}.

Use it when something changes

Keep the baseline. Recheck what changed.

Connect the free comparison to your own release event or scheduler. Unchanged, current inputs reuse the assessment and retain any existing findings. Changed inputs, an expired assessment, or a new scanner version require a fresh assessment.

git clone https://github.com/jdhart81/viridis-agent-fleet.git
cd viridis-agent-fleet
python3 scripts/viridis_preflight_watch.py \
  --base-url https://mcp.viridis-security.com \
  --inputs examples/security-preflight-inputs.json

The example returns BASELINE_REQUIRED and quote instructions. It does not create a scan or payment. Replace the example with your current inputs, then supply the saved paid result on later checks.

Read the complete change-check recipe →

For agent builders

Choose your existing client.

MCP endpoint:

https://mcp.viridis-security.com/security-preflight/mcp

Paid HTTP endpoint:

POST /x402/security-preflight/security_preflight

OpenAPI · Machine-readable service · Agent guide

Before you authorize a purchase

Choose a real manifest and the policy it should satisfy.
Send the intended paid request without a signature to inspect its HTTP 402 quote.
Apply your own budget and payment mandate before signing.
Inspect the result, retain the receipt, and recheck after a relevant change.

The free change check is not a managed monitoring subscription. A failed assessment remains failed when unchanged.

Need help making a whole workflow reliable?

The Agent Reliability Sprint is a separately scoped implementation service for one existing workflow. Security Preflight is the bounded artifact check you can integrate directly.

Verify evidence before relying on it

Use the offline manifest receipt verifier with an operator-approved signing key and scanner. It checks expiry and exact input binding. A verified preflight does not grant tool execution authority or certify runtime safety.

Pay for one static scan by card

Seller: Viridis North LLC (Montana). The ordinary Stripe option costs $1 USD plus applicable tax for one deterministic static scan. Buyers must be in the 50 US states or DC; territories and international buyers are not supported by this card launch. Managed Payments is not used.

Connect your MCP client to https://mcp.viridisconservation.com/payments/mcp. Checkout and fulfillment require your existing fleet caller credential: Authorization: Bearer ... and your registered X-Viridis-Agent-ID. Do not put credentials or card details in scan input.

Before paying: confirm with Justin through your existing arrangement that your caller credential and registered actor are provisioned for checkout and redemption. There is no self-service signup in this flow. Installing a plugin, listing tools, or paying does not provision access. The registered actor header identifies the caller; the intake agent_id identifies the scan subject and does not register an account.

Call create_service_checkout with your actual billing state and intended software use:

{"offer_id":"security-preflight-static-v1","intake":{"action":"scan_source","agent_id":"your-agent","source":"your source text"},"buyer_location":{"country":"US","state":"MT","accepted":true},"buyer_use":{"use":"personal","accepted":true}}

Replace MT with your actual state/DC; use business when that describes your intended use. This declaration is not a tax exemption. Open the returned Stripe URL and enter the same billing location. Retain the exact intake and session ID privately. After paying, call fulfill_paid_scan with session_id and scan_input equal to the original intake.

Only verified payment authorizes work. Delayed payment can be checked again using the same session/input. Paid replay returns the recorded result. A changed billing location, failed work, or interruption is held for owner review; no automatic refund or completion-time promise is made. Findings and the signed receipt do not certify runtime security. Manual report sales remain closed and existing free report commitments remain with Justin.

Keep the session ID private: redemption also requires caller authorization and the same registered caller that created checkout. Direct MCP assessment tools require that verified fixed-order session and exact intake; shared credits are not accepted by this workflow. Legacy orders without a recorded caller require owner review.

Public receipt reads return an unsigned filtered summary with an opaque ID, allowlisted verdict/counts and hashes of the original records. Subject IDs, input digests, detailed findings, tool/policy names, scanner metadata and original signatures are excluded. Retain your original signed delivery for signature verification; the filtered summary does not replace it. Original records can repeat submitted tool and policy names verbatim. Do not submit secrets. The scan order stores the input digest, billing admission, payment evidence and fulfillment result rather than raw scan intake. These boundaries do not promise anonymity or a deletion period.