MCP-native secrets for Cursor

Cursor secrets management
without the risk

Every time you paste an API key into Cursor, it lands in prompts, logs, and LLM context windows. 1claw's MCP server injects secrets at runtime so they never touch your editor.

Start freeInstall the MCP server
The problem

How secrets leak in Cursor

Cursor's AI features are powerful — but every prompt you send includes context from your workspace. When API keys live in .env files, hardcoded strings, or clipboard pastes, they become part of that context. Here's where secrets end up:

  • LLM context windows. Cursor sends open files as context. If your .env is open or referenced, the LLM provider receives your keys.
  • Git history. Even if you add .env to .gitignore, a single accidental commit exposes keys forever. GitHub's secret scanning finds thousands of leaked keys daily.
  • Clipboard and prompts. Pasting a key into a Cursor chat message means it's now in the conversation log, synced to the cloud, and potentially used as training data.
  • Shared workspaces. Team members with Cursor access can see every file. Without granular access control, every developer has every secret.
.env — exposed in workspacein LLM context

# Cursor sends this as context

OPENAI_API_KEY=sk-proj-4f8a...c2d1

STRIPE_SECRET=sk_live_51N8x...kJ2m

DATABASE_URL=postgresql://admin:p@ss@prod:5432

Every key in this file is now in the LLM provider's logs.
The solution

MCP server that plugs into Cursor natively

1claw's MCP server connects to Cursor's Model Context Protocol. Your agent fetches secrets at runtime through authenticated, scoped API calls — never from flat files.

HSM-encrypted vault

Secrets are stored in a hardware-backed vault with AES-256 envelope encryption. Not in .env, not in memory, not in the repo.

Policy-scoped access

Each agent only sees the secret paths you explicitly grant. Glob-based policies, IP restrictions, and time-window conditions.

Zero-config MCP integration

Add the MCP config to your Cursor settings. Your agent gets 17 tools: list, get, put, rotate, share, simulate, sign, and more — all authenticated.

Set up in under two minutes

Add the 1claw MCP server to your Cursor configuration. Once connected, your agent can fetch secrets on demand — no .env files needed. The agent authenticates with an API key and only accesses the secret paths you've granted via policies.

  • Create a vault and store your secrets
  • Register an agent and grant it a policy on the paths it needs
  • Add the MCP config to Cursor — agent fetches secrets at runtime
  • Secrets never appear in your workspace, git history, or prompts
.cursor/mcp.json
{
  "mcpServers": {
    "1claw": {
      "command": "npx",
      "args": ["@1claw/mcp"],
      "env": {
        "ONECLAW_AGENT_API_KEY": "ocv_your_key_here"
      }
    }
  }
}

How it works under the hood

The MCP server handles authentication, token refresh, and vault discovery automatically. Your agent never stores credentials locally.

1

Agent starts

Cursor launches the MCP server process. It exchanges the API key for a short-lived JWT.

2

Vault auto-discovered

The token response includes the agent’s vault IDs. No vault ID configuration needed.

3

Secret fetched

When the agent calls get_secret, the MCP server fetches the value from the encrypted vault over HTTPS.

4

Token refreshed

JWTs auto-refresh 60 seconds before expiry. The agent key stays in the MCP process — never in Cursor memory.

.env files vs. 1claw vault

Flat files were never designed for AI-assisted development.

.env files
1claw vault
Encrypted at rest
—
Access policies
—
Audit trail
—
Visible in LLM context
—
Risk of git commit
—
Secret rotation
—
MCP-native
—
Free tier — 3 vaults, 50 secrets, 1,000 requests/mo

Stop pasting secrets into Cursor

Install the MCP server, create a vault, and your Cursor agent will never see a raw API key again. Setup takes under two minutes.

Start freeRead the MCP docs

Already using 1claw? Jump to the MCP setup guide →