Works with the AI Agents you already use
Miftah — Multi-Account MCP Profiles
Use one MCP connector with several accounts — work, personal, client, production — and let your agent switch deliberately, with credentials kept out of client config.
Free
Miftah — Multi-Account MCP Profiles
Example session with this skill installed
Set up GitHub so I can use my work and personal accounts from one connector in Claude Desktop.
- Read your context and instructions
- Compiled the miftah
Summary
Miftah 1.1.6 is installed. I created one github connector with work and personal profiles. It validates, but doctor correctly fails until you supply the two token references. I did not ask for or handle any token.
miftah init github --preset github --active-profile-lifetime workspace --output ~/.config/miftah/github.json --client claude-desktop
miftah validate --config ~/.config/miftah/github.json
{"ok":true,"name":"github","profiles":["work","personal"]}
Setup report
| Item | Value |
|---|---|
| Config file | ~/.config/miftah/github.json |
| Profiles | work, personal |
| Profile switch | durable (workspace), survives restarts |
| Upstream | GitHub's official MCP server in Docker, read-only toolsets |
| References you must fill | GITHUB_WORK_TOKEN, GITHUB_PERSONAL_TOKEN (names only) |
Checks
validate: passdoctor:
failed (expected right now): configuration, redaction and file permissions pass; secret references for both profiles cannot be resolved yet.
test-profile: not run; it needs the tokens first.
Credentials (you do this)
Claude Desktop does not read ~/.zshrc, so exported variables usually won't reach it. Store each token in your OS keychain or 1Password and point the profile at it (secretref:keychain://... or secretref:op://...). Use least-privilege tokens: Miftah's local policy cannot remove permissions a token already has.
Client snippet
Merge this one entry into mcpServers via Claude Desktop → Developer → Edit Config. Keep the absolute paths as they are.
{
"mcpServers": {
"github": {
"command": "/usr/local/bin/node",
"args": ["/usr/local/lib/node_modules/@lubab/miftah/dist/cli/main.js", "serve", "--config", "~/.config/miftah/github.json"]
}
}
}
Next steps
- Add the two token references, then tell me and I'll rerun
doctorandtest-profile --profile work. - Restart Claude Desktop.
- Ask: "List the profiles available through github" — I'll confirm the active profile before any write.
Connects securely to your tools. The creator never sees your data.
About this skill
Miftah — the right account, every time
You already trust an MCP server for GitHub, Sentry, or Google Search Console. The trouble starts when the same server needs two accounts: you end up with github-work and github-personal as near-identical client entries, tokens pasted into JSON, and no clear signal which account the agent is about to use.
Miftah runs locally in front of that upstream server. Your client sees one connector. Behind it sit named profiles, each with its own credential reference, and the agent lists, checks, and switches them on purpose.
What this skill teaches your agent
- Pick the right setup path. Terminal wizard, browser Console, or a scripted preset (GitHub, Sentry, pinned npx, Docker by digest, remote HTTPS, reviewed local executable).
- Never touch your secrets. It will not ask you to paste tokens, and it will not read or print secret values. It tells you which references to fill: environment variable, OS keychain, 1Password, or dotenv.
- Validate before connecting. Runs
validate,doctor, andtest-profile, and reports results without overstating them. - Connect your client safely. Shows you the exact snippet and where it goes; Miftah never edits client settings silently.
- Confirm the account before writes. Checks the active profile before any change, asks you before switching, and only approves pending actions you approved.
Local by design
- No cloud service and no telemetry.
- Configs hold secret references only, never values.
- Redacted diagnostics and optional hash-chained local audit metadata.
Who it's for
Developers, freelancers, and agencies using one MCP service across work, personal, client, or production accounts.
When it won't help
If you use one account and need none of these controls, keep your direct MCP entry. Miftah wraps an upstream; it does not fix an upstream that doesn't work on its own, and its policy cannot reduce what a provider token is already allowed to do.
Open source (MIT): https://github.com/mohanagy/miftah
Requires Node.js 20+. No support for generic/npx presets on Windows. Local policy cannot override upstream token permissions. Secret values must be provided by the user via external providers.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
File Scopes
Terminal: runs the miftah CLI (init, validate, doctor, test-profile, profile commands). Network: only for the one-time, user-approved `npm install -g @lubab/miftah@1.1.6`; Miftah itself has no cloud service or telemetry, and upstream MCP traffic is the user's own. Read files: only the generated Miftah config, which holds secret references, never values; dotenv files, keychains and token caches are off-limits. Write files: `miftah init` writes ~/.config/miftah/<name>.json; client settings are never edited without the user's approval.