mcpField
K
Keysrs

Keysrs

Local
keysrs.comSecurity1Updated 1d ago

About Keysrs

Keysrs is a Mac API-key vault for agents, keeping real credentials in the macOS Keychain while issuing scoped, short-lived tokens. Its MCP server supports agent workflows for finding, importing, granting, and revoking key access, with a gateway for cloud VMs.

Key features

  1. 01Store real keys in macOS Keychain
  2. 02Issue short-lived, scoped access tokens
  3. 03Approve grants with Touch ID
  4. 04Revoke grants and project tokens
  5. 05Track gateway calls and estimated spend

Use cases

  1. 01Move project API keys into the Keychain
  2. 02Give an agent time-limited access to one key and host
  3. 03Provide cloud VMs gateway access without sharing real keys
  4. 04Review requests and estimated spend by provider

MCP support

Keysrs exposes its key discovery, import, access grant, and revocation workflows through an MCP server that agents connect to. The server requires Keysrs.app.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository
Packages
mcpbKeysrs.mcpb

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Keysrs

Short-lived tokens for AI agents, on your Mac and on cloud VMs. Your real API keys stay in the macOS Keychain; agents get a scoped token instead.

  • MCP-first. Agents talk to Keysrs over MCP: they find keys in a project, import them, and ask you for access. You approve each grant with Touch ID.
  • Short-lived tokens. A grant covers one key, one host and minutes to hours; a project token lasts 30 days. Revoke everything in one call.
  • Cloud VMs too. A VM or sandbox gets a token, never the key. Its calls come back to the gateway on your Mac, over SSH or a relay you deploy yourself, and the real key is swapped in there.

Keysrs also charts the calls made through its gateway: tokens, requests and estimated spend per key and provider.

Free and MIT-licensed. macOS 14+ on Apple Silicon.

Install

Download Keysrs-arm64.dmg from keysrs.com or the latest release, drag Keysrs to Applications and open it. It runs in the background (no Dock icon or window until you open it again) and updates itself.

For the keys command line, choose Install Command Line Tool… in the Keysrs menu, which links ~/.local/bin/keys.

Use it with your agents

keys setup mcp

This registers the Keysrs MCP server with Claude Code, Codex, Cursor, Droid and Claude Desktop. Restart the agent, then ask it in plain words:

  • "Scan this project for API keys with Keysrs." It finds credential-shaped variables in your env files and reports names and last-4 hints, never values.
  • "Import and wrap this project's keys." One Touch ID moves the keys into the macOS Keychain and rewrites the env files to a 30-day project token.
  • "Ask me for a Keysrs grant for the key anthropic, 30 minutes." You approve with Touch ID; the agent gets a token scoped to one key, one host and a time limit.
  • "Revoke everything." Ends every grant, project token and relay pairing at once, and turns the relay off.

keys unwrap puts the original values back.

Cloud VMs

keys gateway on openai                               # once per key
keys client issue openai --label "my vm" --days 7   # a revocable token for the VM
keys tunnel user@my-vm                               # VM reaches the gateway over SSH

On the VM, point the agent at http://127.0.0.1:12767/openai and use the ksfc_… token as its API key. For a sandbox you can't SSH into, keys relay carries the same gateway through a Cloudflare Worker you deploy into your own account, and keys share "my box" --key openai hands a machine several keys at once through it (one token, one pairing, printed as env lines to paste). See cloud agents and the relay.

From the command line

keys add                                  # guided: provider, name, hidden secret
keys list
keys grant openai --minutes 30            # a scoped token for a script
keys env openai OPENAI_API_KEY -- python run.py
keys spend --week
keys help                                 # every command

Details: the full guide, agents and MCP, cloud agents and the relay.

Privacy

  • Everything listens on 127.0.0.1 only. The one optional way in from outside is a relay Worker you deploy into your own Cloudflare account, and keys never travel over it.
  • No tool, log, screen or report ever shows a key value. Copy, reveal, env and rotate ask for Touch ID every time.
  • Keysrs counts only the calls made through its own gateway; it reads no other tool's logs. It never reads ~/.codex/auth.json, ~/.grok/auth.json, browser cookies or provider websites.
  • The app sends no analytics or telemetry.
  • Update checks are one cookieless request to GitHub's release API.

Build from source

make build                     # signed debug CLI at .build/debug/keys
python3 scripts/build-app.py   # Keysrs.app from the release build
swift test                     # offline; CI's env is in test.yml

Signing setup is in SIGNING.md; .github/workflows/test.yml is the full list of checks CI runs.

Keysrs was called Keysreallysafe; the repository, the Keychain service and the data folder keep that name so existing installs keep working.

License

MIT. See LICENSE and third-party notices.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.