Profile the live MCP surface
Inventory tools, prompts, resources, schemas, authentication signals, retrieval behavior, sensitive fields, and capability risk.
MCP Assurance Intelligence
Turn a live MCP surface into reviewed, executable coverage.
Sammy discovers what your MCP server exposes, locks a deterministic testing minimum, proposes missing scenarios, and sends every model-assisted addition through an independent Reviewer and strict validation before a human can approve the Gate.
Deterministic minimumFast · Deep · SpecialistIndependent ReviewerSigned Gates
ILLUSTRATIVE EXAMPLE · Target surfaceLive discovery
Surface 01Tools + schemas
Surface 02Prompts + inputs
Surface 03Resources + RAG
observed evidence
Locked deterministic minimumCannot be weakened by AI
SECURITYCONTRACTSRAGPROMPTSPERFORMANCE
FastFocused coverage
DeepBroad analysis
SpecialistSecurity depth
Independent ReviewerAdd · modify · remove
Deterministic validatorSchema · bindings · lineage
Human approvalImmutable Suite + Policy → signed Gate
3planning modes
ILLUSTRATIVE EXAMPLE · Intelligence depth
3 analysis depths. One validation contract.
Fast, Deep, and Specialist change the planning emphasis—not the trust boundary. Every route receives the same observed surface, keeps the deterministic minimum locked, passes through an independent Reviewer, and must survive deterministic validation.
FastFocused · low latency
DeepBroad · multi-angle
SpecialistSecurity · adversarial
Separate model familyIndependent Reviewer
Shared hard boundaryDeterministic validation
1locked deterministic minimum
1independent Reviewer
0automatic approvals
Honest outcomes. If no model proposal survives validation, Prooflane reports a model-assisted deterministic result instead of claiming expanded coverage.
Coverage planning
MCP quality assurance starts with what the target actually exposes.
Sammy keeps observed surface evidence separate from inference, then converts eligible recommendations into versioned tests with explicit inputs, assertions, and lineage.
Lock mandatory coverage
Five deterministic recommenders establish the controls that model assistance cannot silently remove or weaken.
Choose the right depth
Fast, Deep, and Specialist routes propose bounded scenarios for the connected target and selected assurance needs.
Challenge model proposals
A separate Reviewer can add, modify, or remove non-mandatory coverage before deterministic validation.
Reject weak or unsafe plans
Unknown capabilities, invalid schemas, unresolved template values, broken bindings, missing assertions, and untrusted inputs fail closed.
Approve immutable Gates
Human review, an exact Policy pin, target fingerprint, signed versions, and executable-adapter preflight protect CI.
One governed workflow
Discover, plan, review, and run without lowering the evidence bar.
Model intelligence expands the plan; deterministic engines and signed governance decide what can execute.
Discover
Connect the MCP target and fingerprint its current capability surface.
Plan
Build the mandatory minimum and request bounded model-assisted additions.
Review
Inspect Reviewer diffs, validation interventions, deferred work, and native drafts.
Gate
Approve the exact Suite and Policy, then run the signed Gate from UI or CI.
npx --yes @prooflane/cli@latest gate "$MCP_CONFIG" \ --gate production \ --github
MCP test automation
One plan across the testing jobs teams usually manage separately.
Sammy connects each recommendation to a native Prooflane pillar and leaves cases deferred when the required trusted input or execution adapter does not exist.
Contract and schema testing
Plan capability-drift, request-schema, response-shape, and expected-output coverage against approved captures.
Security test automation
Cover authorization edges, destructive tools, data exposure, prompt injection, unsafe chains, and expected denials.
Automated behavior coverage
Turn prompt and tool-use expectations into repeatable positive, negative, boundary, and regression cases.
Retrieval assurance
Recommend retrieval quality, grounding, poisoning, cross-tenant, contract, and resilience tests when evidence supports them.
Operational confidence
Plan latency, reliability, retry, and budget controls around the MCP calls the application depends on.
CI-ready Gate execution
Run the same approved Suite and Policy from the local UI or GitHub Actions and retain compact governed history.
Governed learning
Turn reviewed run history into the next governed test proposal.
Sammy's learning mechanism is downstream of execution. It uses approved compact non-PASS history to propose a Generation-N DRAFT with exact lineage—never an automatic policy change, and never model training on raw customer evidence.
Completed Gate Run
Only compact REVIEW or BLOCK pillar summaries from a signed, completed Run are eligible.
RUN · GATE · SUITE · POLICYHuman disposition
An attributed decision marks the exact finding set confirmed, remediated, false positive, or accepted risk.
ORIGINAL DECISION PRESERVEDPrivate planner
The Control Plane maps approved history to one unambiguous signed Suite control.
COMPACT HISTORY ONLYGeneration-N DRAFT
Propose a regression, evolution, or eligible suppression-validation test with exact predecessor lineage.
NO AUTOMATIC PROMOTIONDistinct approval
Creator, reviewer, and approver remain separate identities before any learned test becomes approved.
REVIEWED · APPROVEDExact lineageRun → Gate → Suite → Policy → target fingerprint → evidence fingerprint → disposition → learned-test proposal
No raw evidence enters learning
No prompts, responses, retrieved documents, tool traffic, credentials, provider configuration, or local paths.
DRAFT first, every time
Learning can propose the next test; it cannot weaken the locked minimum or silently change an active Gate.
Execution never waits on learning
Gate results, reports, synchronization, and history remain available when learning is disabled or unavailable.
Privacy boundary
Hosted planning without moving raw execution evidence.
Stays with the local Runner
- MCP credentials, configuration, and target connection
- Raw prompts, tool inputs and outputs, and retrieved documents
- Full local test evidence and provider credentials
Governed by the Control Plane
- Sanitized capability metadata for bounded planning
- Signed Gate, Suite, Policy, and public verification identities
- Compact run counts, decisions, timings, and evidence fingerprints
Customer-controlled
Local Runner
Credentials · target calls · raw prompts · tool output · retrieved documents · full evidence
Sanitized metadata + compact evidence
Bounded plan + signed artifacts
Prooflane-hosted
Control Plane
Planning routes · independent review · governance · signatures · compact run history
Continue through the spine
Turn planned coverage into pillar-native evidence.
From discovery to proof
Build the assurance plan your MCP release actually needs.
Keep the deterministic minimum, add reviewed intelligence, and run one signed Gate everywhere.
30-day full-access trial. No credit card required.