Skip to main content
Glama

Invock

Invock is a zero-trust execution and privacy layer that controls, contains, and cryptographically proves AI-agent actions before they reach tools or external systems.

Invock acts as an inline reference monitor between autonomous AI agents (such as Claude Code or OpenAI Codex) and external execution environments (MCP servers, system shells, file systems, APIs). It guarantees that no action executes without explicit policy authorization, valid identity bindings, bounded capability leases, and strict privacy mode compliance.


The Problem

Modern AI coding agents are granted powerful system access to read codebases, run shell commands, invoke APIs, and manage databases. However:

  • Prompt instructions are not security boundaries: Prompt injections or model hallucination can cause agents to exceed intended operational scopes.

  • Unbounded tool execution: Standard MCP or SDK tool calls carry side effects that can destroy local state, leak sensitive keys, or perform unauthorized network egress.

  • Historical data exposure: Installing a protective agent tool today does not retroactively clean pre-existing sensitive artifacts, conversation caches, or external provider logs created prior to installation.


Related MCP server: open-edison

What Invock Does

Invock enforces a strict fail-closed, deterministic pre-execution boundary:

  • Deterministic Pre-Execution Authorization: Evaluates tool calls against structured security policies, intent capsules, and delegated capability leases before execution occurs.

  • Claude Code & Codex Integration: Wraps standard CLI clients and MCP transports to intercept and sanitize tool invocations transparently.

  • Signed Receipt Chains: Issues Ed25519-signed cryptographic receipts for every authorization decision and execution outcome.

  • Local & End-to-End Zero Data Retention (ZDR): Guarantees content-free local persistence (LOCAL_ZDR) and blocks unverified third-party data routes (END_TO_END_ZDR).

  • Legacy Privacy Remediation & Protection Boundary: Discovers pre-installation privacy risks, enables safe local artifact removal, provides provider-history guidance, and issues signed protection boundary certificates.

  • Local Dashboard & Visual Telemetry: Exposes content-free real-time monitoring and approval interfaces over a local loopback server.


Quick Start

Prerequisites

  • Node.js >=22.5.0

  • pnpm 11.15.1

Installation & Initialization

# Install dependencies and build project
pnpm install --frozen-lockfile
pnpm build

# Initialize local Invock control plane
pnpm invock init

# Run safe legacy privacy onboarding
pnpm invock privacy onboard --yes

Five-Minute Demo

Experience Invock's protection layer and legacy privacy onboarding in under five minutes:

# 1. Run the interactive safe policy demo
pnpm invock demo:safe

# 2. Run the adversarial attack prevention demo
pnpm invock demo:attack

# 3. Run the synthetic legacy privacy onboarding demo
pnpm invock privacy legacy demo

# 4. View real-time content-free stats
pnpm invock stats

Everyday Usage

Agent Wrapping & Protection

Wrap your preferred AI coding agent CLI to enforce Invock policy mediation automatically:

# Launch Claude Code under Invock protection
pnpm invock wrap claude -- --help

# Launch Codex under Invock protection
pnpm invock wrap codex -- --help

Legacy Privacy Commands

Inspect and manage historical data risks created before Invock was installed:

# Check onboarding and privacy status
pnpm invock privacy legacy status

# Run local legacy artifact scan
pnpm invock privacy legacy scan --yes

# View external provider history guidance
pnpm invock privacy legacy provider-actions

# Inspect signed Protection Boundary certificate
pnpm invock privacy boundary show

# Verify Protection Boundary signature
pnpm invock privacy boundary verify

Architecture

  Claude Code / Codex / MCP Client
                 |
                 v
            Invock CLI
                 |
  +--------------+--------------+
  |                             |
  v                             v
Policy Gate               Privacy Engine
(Capsules, Leases,       (Local ZDR, E2E ZDR,
 Policy Matrix)           Legacy Scanner)
  |                             |
  +--------------+--------------+
                 |
                 v
       Execution / Isolation
   (Contained Runner, MCP Stdio)
                 |
                 v
 Signed Ed25519 Receipt Chain & Store

Request Lifecycle

  1. Invocation Capture: The agent attempts a tool execution (e.g., file read, terminal execution, HTTP request).

  2. Canonical Normalization: The raw tool call is converted into a canonical ActionEnvelope.

  3. Identity & Context Binding: Active workload identity, project context, and session parameters are verified.

  4. Policy Evaluation: Invock checks the requested tool and exact arguments against active policy rules.

  5. Privacy Mode Check: Verifies retention profiles against active LOCAL_ZDR or END_TO_END_ZDR constraints.

  6. Authorization Decision: Returns ALLOW, BLOCK, or APPROVAL_REQUIRED.

  7. Execution & Receipt Signing: If allowed, execution occurs within a contained runner and a signed Ed25519 receipt is recorded in the immutable local store.


Security Model

  • Pre-Execution Control Point: Authorization occurs prior to side effects.

  • Fail-Closed Default: Mismatched signatures, unhandled schema drift, or unknown capabilities immediately yield a BLOCK verdict.

  • Exact Argument Authorization: Policies bind to canonicalized argument values, preventing path traversal or parameter tampering.

  • Cryptographic Attestation: Software identities and receipts are signed using Ed25519 key pairs.

  • Honest Mediation Boundary: Invock strictly controls actions visible at its supported local proxy and gateway interfaces.

Note: Invock controls actions visible at its supported mediation and containment boundaries. It does not prove that external infrastructure or an allowed remote service behaves honestly beyond those boundaries.


Privacy

Invock implements a multi-tier privacy framework designed to enforce Zero Data Retention (ZDR):

Local ZDR

Ensures that no customer prompt content, code snippets, or tool arguments are written to local logs or persistent databases. Only content-free hashes, timestamps, and reason codes are retained.

End-to-End ZDR

Evaluates downstream processor profiles before request forwarding. If a downstream provider retains customer data or lacks a verified retention profile, invocation is blocked.

Legacy Privacy Onboarding

Discovers legacy local history (Claude logs, Codex caches, old Invock artifacts) created before Invock installation, allowing safe user-confirmed removal of disposable files.

Protection Boundary

Issues an Ed25519-signed receipt certifying the exact timestamp when Invock protection became active, providing cryptographic proof of boundary enforcement without claiming false retroactive erasure of external provider logs.


Core Command Reference

Command

Description

Example

invock init

Initialize local control plane state

pnpm invock init

invock status

Inspect CLI state and integrations

pnpm invock status

invock privacy status

Check active privacy mode and contract

pnpm invock privacy status

invock privacy onboard

Run guided legacy privacy onboarding

pnpm invock privacy onboard --yes

invock privacy legacy status

View legacy audit status and findings

pnpm invock privacy legacy status

invock privacy boundary show

Display active signed protection boundary

pnpm invock privacy boundary show

invock privacy boundary verify

Verify cryptographic boundary signature

pnpm invock privacy boundary verify

invock privacy legacy demo

Run synthetic onboarding demonstration

pnpm invock privacy legacy demo

invock wrap <agent>

Launch agent binary under Invock protection

pnpm invock wrap claude

invock stats

Output content-free telemetry statistics

pnpm invock stats --json


Configuration

Invock defaults to secure, local-only defaults stored in .invock/:

  • INVOCK_HOME: Base directory for local state (defaults to ~/.invock).

  • INVOCK_PRIVACY_DIR: Custom directory for privacy configurations and boundaries.

  • Loopback Binding: Local server and proxy bind strictly to 127.0.0.1.

  • Zero Content Logging: Logging of sensitive prompts and tool outputs is disabled by default.


Supported Integrations

Integration

Status

Usage

Limitations

Claude Code

Supported

pnpm invock wrap claude

Requires local Claude CLI installation

OpenAI Codex

Supported

pnpm invock wrap codex

Requires local Codex CLI installation

Cursor

Preview

pnpm invock wrap cursor

Experimental wrapper; live verification pending


Receipts and Verification

Every authorization decision produces a tamper-evident, Ed25519-signed receipt:

  • Ed25519 Signature: Covers canonical receipt metadata and policy digests.

  • Content-Free Telemetry: Receipts store SHA-256 hashes of actions rather than raw data.

  • Verification: Verify receipt authenticity via pnpm certify or the local dashboard API.


Dashboard

Invock includes a local, loopback-only visual dashboard for monitoring activity:

pnpm invock dashboard
  • Displays real-time content-free invocation statistics.

  • Shows active policy rules, privacy mode status, and receipt digests.

  • Runs entirely locally without external telemetry.


Development & Testing

Run the full validation suite:

# Typecheck TypeScript sources
pnpm typecheck

# Run unit and integration tests (322 tests)
pnpm test

# Build production bundle
pnpm build

# Execute security, CLI, and ZDR certification suites
pnpm certify
pnpm cli:certify
pnpm zdr:certify
pnpm zdr:audit

Repository Structure

src/
├── authority/      # Intent capsules and capability leases
├── containment/    # Contained execution runner and isolation
├── core/           # Canonical serialization and policy engine
├── evidence/       # Content-free telemetry and evidence bundles
├── gateway/        # MCP stdio & HTTP invocation gates
├── identity/       # Workload identity authority
├── privacy/        # ZDR engine and Legacy Privacy remediation
└── storage/        # Ed25519 signed receipt store & SQLite DB
scripts/            # Validation, certification, and audit scripts
test/               # Unit, integration, and security test suite

Troubleshooting

Invock Command Not Found

Ensure the project is built via pnpm build or run commands using pnpm invock <command>.

Agent Not Detected

Ensure claude or codex binaries are installed and accessible in your system PATH.

End-to-End ZDR Blocked

If END_TO_END_ZDR returns BLOCK, an unlisted or non-compliant AI processor was detected in the execution route. Register verified retention profiles via pnpm invock privacy processors add.


Honest Limitations

  • Supported Mediation Boundaries: Invock mediates tool invocations routed through its gateway or wrappers. Unwrapped direct shell calls bypass local interception.

  • Provider-Side Retained Data: Invock can clean local legacy files and issue signed protection boundaries, but cannot forcibly delete data stored inside third-party cloud provider logs.

  • Memory & Storage: High-frequency tool execution produces receipt records in SQLite; periodic cleanup or pruning may be configured.


License

This project is licensed under the MIT License.

A
license - permissive license
-
quality - not tested
A
maintenance

Maintenance

Maintainers
Response time
Release cycle
1Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

  • A
    license
    -
    quality
    D
    maintenance
    A governance and control layer for MCP tools that manages tool requests as intents through policy-based approval, queuing, or blocking. It enables secure human oversight and audit trails for consequential agent actions across platforms like Claude Desktop and Cursor.
    Last updated
    1
    MIT No Attribution
  • A
    license
    -
    quality
    D
    maintenance
    Deterministic agentic data firewall that secures, monitors, and controls AI agent interactions with data and software via MCP, preventing data leaks.
    Last updated
    287
    GPL 3.0
  • A
    license
    B
    quality
    A
    maintenance
    A governance proxy for AI tools — every MCP/agent tool call is policy-gated, secret-redacted, and written to a hash-chained, offline-verifiable audit trail.
    Last updated
    13
    MIT
  • A
    license
    -
    quality
    C
    maintenance
    A pre-action authorization server for AI agents that classifies tool calls into 14 intent categories, scores risk 0-100, and produces deterministic allow/deny/ask decisions with full audit trail.
    Last updated
    MIT

View all related MCP servers

Related MCP Connectors

  • See, price, and control every tool call your AI agents make: policy checks, cost, and audit tools.

  • Runtime permission, approval, and audit layer for AI agent tool execution.

  • Security firewall for AI agents — scans MCP calls for injection, secrets, and risks.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/saiaathish/Invock'

If you have feedback or need assistance with the MCP directory API, please join our Discord server