Bespoke · From £10,000

Pharaoh Agent Shield

Deterministic runtime MCP security gateway between AI agents and your tools, APIs, and data — enforcing named human owners, task-scoped credential brokerage, schema-drift fingerprinting, secret redaction, hash-bound approvals, CycloneDX release-evidence verification, and a server-side kill switch.

LIVE MCP GOVERNANCE DEMO · AGENT-SHIELD.PHAROAHTECHNOLOGY.CO.UK Open Live Demo Door ↗
Pharaoh Agent Shield live demo console showing ENFORCE mode, registered agents, held human approvals, and verified SHA-256 audit hash chain
Live agent-shield.pharoahtechnology.co.uk governance console demonstrating named agent ownership, task-scoped credential leases, T0–T4 approval thresholds, and UK ICO / EU AI Act Art. 50 alignment.
Overview

What Pharaoh Agent Shield is.

Pharaoh Agent Shield sits deterministically between every AI agent and the MCP servers it can invoke. No agent executes without a registered human owner, a short-lived task- and tenant-bound credential lease (shld_lease_...), a pinned tool-schema fingerprint, secret-redacted arguments, tier-graded human approval (T0–T4), and a reconstructable SHA-256 hash-chained action log.

Capabilities

What it does.

Runtime MCP Firewall & Audit Controls

Named Human Owner Register

Refuses any tool call from an unregistered agent identity (AGENT_OWNER_REQUIRED), binding every autonomous agent to an accountable human owner.

Task-Scoped Credential Broker (Spec §15)

Issues short-lived capability tokens (shld_lease_...) scoped to tenant, agent, session, task, server, and allowed tools, storing only SHA-256 token hashes with immediate individual or session-cascade revocation and secret-free audit attribution.

Schema-Drift & Rug-Pull Fingerprinting

Pins cryptographic hashes of MCP tool descriptions and JSON schemas so a mutated upstream tool server cannot silently inject new instructions.

Single-Use Hash-Bound Approvals

Holds high-impact tool calls for human sign-off bound to the exact request hash, preventing approval replay or parameter substitution.

Adversarial Evaluation & CycloneDX Release Evidence

Includes a repeatable 5-scenario adversarial evaluation runner (prompt-injection exfiltration, cross-tenant escape, revoked lease continuation, approval mutation, and benign false-positive checks) plus deterministic CycloneDX 1.5 SBOM and Ed25519 release-manifest verification.

Architecture

How it's built.

Every MCP tool invocation traverses a synchronous deterministic policy pipeline before reaching the target tool server, backed by a task-scoped credential broker and a persistent server-side kill switch.

Agent Identity & Scoped Lease GateVerifies registered agent owner and activeSHA-256 task-scoped credential lease.MCP Schema Fingerprint GateBlocks tool-description rug-pulls andunauthenticated MCP servers.Tiered Approval & Secret ScrubberEnforces T0–T4 thresholds, single-userequest hashes, and redaction.

Pharaoh Agent Shield architecture, in order

  1. Agent Identity & Scoped Lease Gate: Verifies registered agent owner and active SHA-256 task-scoped credential lease.
  2. MCP Schema Fingerprint Gate: Blocks tool-description rug-pulls and unauthenticated MCP servers.
  3. Tiered Approval & Secret Scrubber: Enforces T0–T4 thresholds, single-use request hashes, and redaction.
Use Cases

Who it's built for.

Enterprise engineering & internal AI copilots

Allow coding and ops agents to read telemetry freely under task-scoped credential leases while holding production database writes, merges, or customer sends behind single-use human approval.

UK GDPR, EU AI Act & EU CRA release readiness

Maintain exportable, secret-redacted SHA-256 audit trails and Ed25519-signed CycloneDX 1.5 SBOM release bundles proving human oversight and software supply-chain integrity.

FAQ

Straight answers.

Can I inspect Pharaoh Agent Shield live before requesting a quote?

Yes. The isolated demo tenant at agent-shield.pharoahtechnology.co.uk lets you inspect live ENFORCE mode posture, approval queues, and exportable SHA-256 action logs.

How does the Agent Shield Credential Broker prevent credential leakage or reuse?

Agents receive short-lived capability tokens (shld_lease_...) bound to a specific tenant, session, task, MCP server, and tool allowlist. Only the SHA-256 hash is stored in SQLite, upstream secrets are resolved inside the gateway boundary, and revoking a lease or session immediately blocks further calls while recording leaseId and taskId in the hash-chained audit trail without plaintext secrets.

How is Pharaoh Agent Shield deployed and priced?

Bespoke single-tenant cloud or on-premises deployments start from £10,000 by written quotation.

Engagement

How to work with us.

Pharaoh Agent Shield is deployed onto your own Cloudflare, cloud VPC, or on-premises infrastructure and wired around your MCP servers, secret vault, and identity provider.

Bespoke deployments start from £10,000 by written quotation.

Ready to talk about Pharaoh Agent Shield?

Deterministic runtime MCP security gateway between AI agents and your tools, APIs, and data — enforcing named human owners, task-scoped credential brokerage, schema-drift fingerprinting, secret redaction, hash-bound approvals, CycloneDX release-evidence verification, and a server-side kill switch.