This guide provides comprehensive instructions for installing, configuring, and using the WebCrypt Model Context Protocol (MCP) Server across all major AI code editors and agent environments.
- Why Use WebCrypt MCP? (Key Benefits)
- Automatic Zero-Configuration Setup
- Google Antigravity
- Cursor IDE
- Claude Desktop
- VS Code / GitHub Copilot Agent
- Windsurf / Codeium Cascade
- Cline / Roo Code
- Zed Editor
- Health Diagnostics & Verification (
doctor) - Global Multi-Project Management
- Agent Skill & Tool Calling Reference
AI coding agents routinely handle sensitive material: API keys, database credentials, architecture plans, and multi-agent conversation logs.
WebCrypt MCP equips agents with an autonomous, zero-dependency cryptographic vault directly in their toolbelt:
- 🔐 Confidential Local Vaulting: Encrypts credentials and state memory before writing to disk, preventing plaintext leaks in prompt histories and git logs.
- 🛡️ Cryptographic Provenance: Signs release outputs, code refactors, and test evidence packs using standard ECDSA / HMAC digital signatures.
- 🤝 Inter-Agent Security: Generates ephemeral JWK keypairs (RSA-4096 / ECDH) for end-to-end encrypted subagent communication.
- 🧠 PuterVision Triad Standard: Seamless synergy with
state-memory-mcp(workflow DAG) andvision-memory-mcp(visual cache). - ⚡ Zero Native Dependencies: 100% W3C Web Crypto API (
crypto.subtle) execution across Node.js 18+, Bun, browsers, and edge environments. - ⚛️ Post-Quantum Ready: Forward-secure Kyber KEM and Dilithium signatures.
The fastest way to configure WebCrypt for any project is to install it globally and run webcrypt init:
# 1. Install globally
npm install -g webcrypt
# 2. Run init inside your project root
cd /path/to/my-project
webcrypt init(Or via npx without global install: npx -y webcrypt init)
- Registers Project: Saves project path into global registry
~/.webcrypt/projects.json. - Installs Agent Skill: Generates
.agents/skills/webcrypt-mcp/SKILL.md. - Scaffolds MCP Configurations: Creates/updates
.cursor/mcp.jsonand.vscode/mcp.json. - Configures Global Antigravity: Merges MCP server into
~/.gemini/config/config.json. - Injects Agent Workflow Rules: Adds standard
<!-- webcrypt-mcp:start -->blocks to.agents/AGENTS.md,.cursorrules,.windsurfrules,.gemini/instructions.md,.github/copilot-instructions.md, andCLAUDE.md.
Google Antigravity automatically discovers MCP servers defined in global configuration or local workspaces.
Add webcrypt under mcpServers:
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}To allow Antigravity to run cryptographic operations seamlessly without prompting:
{
"globalPermissionGrants": {
"allow": ["command(webcrypt-mcp)", "command(npx webcrypt mcp)"]
}
}- Open Antigravity and check the MCP servers list.
webcryptwill be listed with 6 active tools:encrypt_payload,decrypt_payload,manage_keys,crypto_hash,sign_verify,pqc_kem_sign.
Cursor supports project-specific MCP configurations located in .cursor/mcp.json.
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}Ensure your .cursorrules includes the WebCrypt workflow block:
<!-- webcrypt-mcp:start -->
# Cryptographic Vault & Security (webcrypt-mcp)
Whenever saving sensitive credentials, tokens, or private workflow states, encrypt them using `encrypt_payload(mode: "data", password: "...")` or `encrypt_payload(mode: "symmetric", password: "...")`. Use `manage_keys(action: "generate", type: "rsa" | "ecdh" | "hmac")` to generate keys for inter-agent communication.
<!-- webcrypt-mcp:end -->- Navigate to Cursor Settings -> Features -> MCP.
- Verify
webcryptshows a green status dot with 6 tools available.
Claude Desktop connects to MCP servers over standard input/output (stdio).
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}VS Code supports MCP via the official MCP extension or GitHub Copilot Agent mode.
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}<!-- webcrypt-mcp:start -->
# Cryptographic Vault & Security (webcrypt-mcp)
When handling secret tokens or encrypting vault artifacts, invoke `webcrypt` MCP tools.
<!-- webcrypt-mcp:end -->Windsurf stores global MCP registrations in ~/.codeium/windsurf/mcp_config.json.
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}Cline and Roo Code manage MCP servers through cline_mcp_settings.json.
{
"mcpServers": {
"webcrypt": {
"command": "npx",
"args": ["-y", "webcrypt", "mcp"],
"disabled": false,
"autoApprove": [
"encrypt_payload",
"decrypt_payload",
"manage_keys",
"crypto_hash",
"sign_verify",
"pqc_kem_sign"
]
}
}
}Zed supports MCP servers natively through settings.
{
"context_servers": {
"webcrypt": {
"command": {
"path": "npx",
"args": ["-y", "webcrypt", "mcp"]
}
}
}
}To verify that your environment, SubtleCrypto engines, skill files, and MCP configs are functioning properly:
# Check current project health
webcrypt doctor
# Audit all registered projects across your machine
webcrypt doctor-globalExample Doctor Output:
🩺 Running WebCrypt health diagnostics for: /path/to/my-project
✅ Node.js Runtime: v22.22.2 (Supported)
✅ Web Crypto Engine: AES-GCM, PBKDF2, RSA-OAEP, ECDH subtle engine operational
✅ Agent Skill (.agents/skills/webcrypt-mcp/SKILL.md): Installed (3060 bytes)
✅ IDE MCP Server Registration: Configured in .cursor/mcp.json
✅ Agent Instruction Markers: Active across 5 instruction files
Overall Health: ✅ HEALTHY
WebCrypt maintains an index of your initialized projects in ~/.webcrypt/projects.json:
# List all registered projects
webcrypt projects
# Synchronize all projects with the latest skills & instruction updates
webcrypt init-global
# Scan an entire directory tree (e.g. workspace) and auto-register all projects
webcrypt init-global --scan ~/workspaces
# Remove stale / deleted projects from registry
webcrypt doctor-global --clean-staleWhen WebCrypt is active, AI agents can use these 6 tools directly:
| Tool | Action | Example Agent Invocation |
|---|---|---|
encrypt_payload |
Symmetric Encryption | encrypt_payload({ mode: "data", data: { token: "abc" }, password: "pass" }) |
decrypt_payload |
Symmetric Decryption | decrypt_payload({ mode: "data", ciphertext: "...", password: "pass" }) |
manage_keys |
Keypair Generation | manage_keys({ action: "generate", type: "rsa", modulusLength: 4096 }) |
crypto_hash |
Compute Hashes | crypto_hash({ algorithm: "SHA-256", data: "hello", encoding: "hex" }) |
sign_verify |
ECDSA / HMAC Sign | sign_verify({ action: "sign", algorithm: "HMAC", data: "proof", password: "p" }) |
pqc_kem_sign |
Kyber / Hybrid KEM | pqc_kem_sign({ action: "generate_kyber_keypair", level: "Kyber768" }) |