enigmagent-mcp
enigmagent-mcp is a local encrypted vault server for MCP clients that exposes secret metadata by default and can optionally resolve raw secrets when explicitly enabled.
List secret names, bound domains, and creation timestamps via
enigmagent_listwithout revealing values.With
--allow-raw-resolveorENIGMAGENT_ALLOW_RAW_RESOLVE=1, resolve placeholders viaenigmagent_resolve, enforcing origin/domain binding; supports{{NAME}},{{LOGIN:domain}}, and{{DOC:filename}}syntax passed without braces.Serve authenticated REST endpoints on loopback: unauthenticated
/healthliveness, plus token-protected/status,/list, and/resolvewhen raw resolution is enabled.Store secrets encrypted at rest with Argon2id-derived AES-256-GCM, support v1 vault reads, and use backup/recovery for vault writes.
Provide MCP client configuration adapters for Claude Desktop, Cursor, Continue, Cline, Open WebUI, AnythingLLM, LM Studio, Zed, Goose, and Windsurf.
enigmagent-mcp 2.0.0
enigmagent-mcp is a local encrypted vault server for MCP-capable clients. It
stores secret values with an Argon2id-derived AES-256-GCM key and keeps the
vault file encrypted at rest.
Version 2.0.0 makes the security boundary explicit:
MCP starts in metadata-only mode. Raw secret resolution is disabled unless the operator passes
--allow-raw-resolveor setsENIGMAGENT_ALLOW_RAW_RESOLVE=1.REST binds to
127.0.0.1and requires a 32–256 character Bearer token.REST requests have bounded headers, bodies, and timeouts; responses do not include arbitrary exception messages or permissive browser CORS.
Vault writes use a temporary file, flush, replacement, and a recovery backup.
Version 1 vaults remain readable. New vaults use format version 2 and the versioned KDF context
enigma/v2.
Raw resolution is deliberately an opt-in escape hatch. A caller-declared origin is not proof of the network destination, and a trusted MCP client can still pass returned plaintext to its model. Do not enable raw resolution for an untrusted client or a shared process.
Efficiency and practical value. A common local server centralizes encrypted storage and client configuration, reducing the need to implement vault handling separately for each MCP client. Metadata-only defaults, bounded requests and recovery backups make its operational boundary more explicit than ad hoc plaintext configuration. The reusable adapters support setup consistency, subject to the documented trust limits.
Install and run
npx enigmagent-mcp@2.0.0 --vault ./my.vault.jsonThe first run prompts for the existing vault credentials when attached to a
terminal. For a trusted headless process, set ENIGMAGENT_USER and
ENIGMAGENT_PASS in its private environment. The server does not print either
value.
Related MCP server: Vaulted MCP Server
MCP modes
The stdio transport implements bounded JSON-RPC MCP initialization, ping,
tools/list, and tools/call. The default tool is:
enigmagent_list: names, domains, and creation timestamps only.
With explicit raw-resolution opt-in, it additionally exposes:
enigmagent_resolve: returns a value only when the requested origin matches the entry's domain binding.
The origin is normalized and limited to HTTP(S) URLs without embedded user credentials. Notifications, malformed JSON, oversized frames, invalid tool arguments, and calls before initialization are handled without terminating the process.
Authenticated REST
set ENIGMAGENT_API_TOKEN=use-a-random-32-character-token-or-longer
npx enigmagent-mcp@2.0.0 --mode rest --port 3737 --vault ./my.vault.jsonGET /health is a minimal unauthenticated liveness check. GET /status and
GET /list require Authorization: Bearer <token>. POST /resolve also
requires the token, JSON content type, and the explicit raw-resolution flag.
The service rejects browser-origin requests and binds only to loopback.
Client configuration adapters
integrations.js provides tested configuration builders for Claude Desktop,
Cursor, Continue, Cline, Open WebUI, AnythingLLM, LM Studio, Zed, Goose, and
Windsurf. They are maintained configuration examples using the standard MCP
stdio command; they are not upstream plugins, endorsements, or claims of
adoption by those projects.
Development and verification
npm ci
npm test
npm run check
npm run benchmark
npm pack --dry-runThe tests cover encrypted-at-rest behavior, domain matching, wrong credentials, version-1 migration reads, backup recovery, bounded MCP framing and session ordering, authenticated REST, and all ten configuration adapters. The benchmark uses synthetic data and reports create/add/resolve latency without printing a secret.
Release contents
The historical 1.0.5 tree is preserved in versions/1.0.5/ with its source
archive and SHA-256 manifest. The 2.0.0 release is represented consistently in
package.json, server.json, and manifest.json; publication remains subject
to the repository's CI and release checks.
Scope and threat model
This project protects vault contents from ordinary plaintext-at-rest exposure
and from accidental return by the metadata listing tool. It does not protect
against a compromised operating system, a process with access to the unlocked
memory, malicious client code, side channels, swap/core dumps, or an operator
who explicitly enables raw resolution for an untrusted model. Review
SECURITY.md in the main EnigmAgent project for broader ecosystem guidance.
MIT licensed. See LICENSE.
Available Tools
2 toolsenigmagent_listA
List all secret names and their bound domains. Never returns actual secret values.
Inputs: no named parameters.
enigmagent_resolveA
Resolve a secret placeholder from the EnigmAgent vault. Returns the decrypted value. Domain binding is enforced: the origin must match the secret's bound domain.
Inputs: placeholder (required), origin (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
The two tools have completely distinct purposes: one lists secret names and domains, the other resolves placeholder values. There is no ambiguity between them.
Both tools share a consistent 'enigmagent_' prefix followed by a verb ('list', 'resolve'), following a predictable verb_noun pattern.
With only 2 tools, the set is on the low end of what is reasonable. While each tool is distinct, the total number feels minimal for a secret management server.
The tool surface lacks basic CRUD operations like create, update, or delete secrets. Agents can only list and resolve, which is insufficient for comprehensive secret management.
Maintenance
Related MCP Connectors
A secret store for AI agents: the agent never sees the plaintext.
Password vault for AI agents with scoped, user-approved access
Encrypted secret store and rotation for autonomous agent credentials
Secrets for developers and agents—secure context and workflows without exposing secret values.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceLocal encrypted vault for API keys. LLMs never see your real keys — the server injects them transparently into HTTP requests.-
- AlicenseAqualityBmaintenanceShare encrypted, self-destructing secrets from your AI agent. Zero-knowledge E2E encryption. Agent-blind input sources (env:, file:, dotenv:) keep secrets out of LLM context.448 npm1MIT
- AlicenseBqualityBmaintenanceEncrypted secrets vault that blinds AI agents to API keys. Stores secrets in AES-256-GCM encrypted SQLite vault, resolves them at runtime via MCP values never appear in LLM conversation transcripts. Sandbox .env files with deterministic fakes.27133 npm3MIT
- AlicenseNot gradedqualityDmaintenanceAgent Vault is a local credential vault for AI agents. It enables agents to use secrets by name without ever seeing their values, with host allowlists, output scrubbing, and audit logging.4MIT