The problem

If you work with more than one company, you use the same MCP server (Supabase, GitHub, Slack…) with different identities — a different account, email and token per client. Today most LLM clients hold one account at a time per connector: to switch client you disconnect, reconnect, and redo the OAuth login. Every time.

The MCP protocol has no notion of "account": one session = one identity = one set of credentials. JanusMCP fills that gap.

What it does

JanusMCP is a local multi-account tool broker for AI agents. It can be used directly from the CLI or as an MCP server in front of the real upstream servers:

• Add N accounts once for the same service and keep them all available.
• Switch identity without reconnecting — no re-login, no fiddling with config.
• Works with any LLM client — it just speaks standard MCP (stdio + Streamable HTTP).
• Works without an MCP session too — agents can drive it directly from the CLI (code-execution mode): janusmcp tools / schema / call invoke tools on demand, with no bulk tool definitions loaded upfront.
• Runs locally — your machine, your keychain, your control.
• Keeps the context clean — it exposes only the active account's tools, not N×tools.

                 ┌─────────────────────────────┐   ┌─ Supabase (Client A)
LLM client ─MCP─▶│   JanusMCP broker           │─▶ ├─ Supabase (Client B)
(Claude/GPT/     │   active-account · vault ·  │   ├─ GitHub  (Client A)
 Gemini/…)       │   per-session scoping       │   └─ …
                 └─────────────────────────────┘

You drive it with three control tools that appear in any client: janus_list_accounts, janus_use_account, janus_whoami.

Choose your mode

Use from the CLI — Claude Code, Codex, Cursor, scripts

npm install -g @bayway/janusmcp
janusmcp tools
janusmcp schema list_tables
janusmcp call list_tables --args '{"schemas":["public"]}' --json --timeout 30s

The CLI discovers schemas only when needed, accepts JSON through --args or stdin, supports explicit accounts/profiles, and provides stable exit codes. See the CLI guide and copyable agent instructions.

Connect over MCP — desktop and chat clients

npx @bayway/janusmcp serve
# or configure a supported client automatically:
janusmcp install claude-desktop

Both modes share the same config, OS-keychain vault, OAuth tokens and persisted active account. Use janusmcp daemon start to keep upstream sessions warm across repeated CLI calls.

Install

Once released, install via your favorite channel (all published automatically on each tag — see RELEASING.md):

npx @bayway/janusmcp serve                       # npm / MCP mode
brew install bayway/janusmcp/janusmcp     # Homebrew (macOS/Linux)
scoop install janusmcp                               # Windows
docker run --rm -p 7332:7332 ghcr.io/bayway/janusmcp:latest

…or download a prebuilt binary from Releases.

Build from source (60-second quickstart)

git clone https://github.com/bayway/janusmcp
cd janusmcp/go
make build                       # produces ./bin/janusmcp
cp config.example.json config.json   # edit with your accounts
./bin/janusmcp serve             # stdio, for Claude Desktop/Code

Two Supabase clients, PATs kept in your OS keychain (never in the config):

./bin/janusmcp vault set supabase_client_a   # paste the PAT
./bin/janusmcp vault set supabase_client_b

// config.json
{
  "bindingMode": "session",
  "accounts": [
    { "id": "client_a", "service": "supabase", "command": "npx",
      "args": ["-y", "@supabase/mcp-server-supabase@latest", "--read-only", "--project-ref=REF_A"],
      "env": { "SUPABASE_ACCESS_TOKEN": "vault:supabase_client_a" } },
    { "id": "client_b", "service": "supabase", "command": "npx",
      "args": ["-y", "@supabase/mcp-server-supabase@latest", "--read-only", "--project-ref=REF_B"],
      "env": { "SUPABASE_ACCESS_TOKEN": "vault:supabase_client_b" } }
  ]
}

Add it to Claude Desktop:

{ "mcpServers": { "janusmcp": {
  "command": "/abs/path/janusmcp/go/bin/janusmcp", "args": ["serve"],
  "env": { "JANUS_CONFIG": "/abs/path/janusmcp/go/config.json" } } } }

For ChatGPT / Gemini / Cursor / Copilot, run HTTP and point them at the URL:

JANUS_TRANSPORT=http JANUS_HTTP_PORT=7332 ./bin/janusmcp serve
# → http://127.0.0.1:7332/mcp

Commands

Run janusmcp help for the full reference. The essentials:

Supported <client> values for install / uninstall: claude-desktop, claude-code, cursor, vscode, gemini, codex, chatgpt, print. Run janusmcp install list (or uninstall list) to see each target and whether it's already configured.

janusmcp install claude-desktop      # one-command setup
janusmcp uninstall claude-desktop    # clean removal (restart the client afterwards)

Inside any connected client you also get the control tools janus_list_accounts, janus_use_account, janus_whoami, janus_login, janus_use_profile, and janus_with_account.

Profiles — a whole client's stack at once

A profile groups accounts of the same client across different services. Activating it exposes the tools of all its accounts together, and each call is routed to the right upstream:

// config.json
{
  "accounts": [
    { "id": "supabase_a", "service": "supabase", "transport": "http", "url": "https://mcp.supabase.com/mcp", "auth": "oauth" },
    { "id": "github_a",   "service": "github",   "transport": "http", "url": "https://api.githubcopilot.com/mcp/", "auth": "oauth" }
  ],
  "profiles": {
    "client_a": ["supabase_a", "github_a"]
  }
}

Then in chat: janus_use_profile with { "profile": "client_a" } → Supabase and GitHub tools for Client A are available simultaneously. Colliding tool names across accounts are namespaced (<account>_<tool>).

One-shot cross-account calls

janus_with_account runs a single call on another account without changing the active one — e.g. { "account_id": "client_b", "tool": "list_tables" }. Omit tool to list that account's available tools first.

Code-execution mode — context-efficient tools from the terminal

Loading every MCP tool definition into an LLM context is expensive. In code-execution mode an agent (or you) invokes tools on demand from the shell instead — à la "code execution with MCP" — so the context holds only the results it actually asked for:

janusmcp tools                        # compact list for the active account/profile
janusmcp tools client_a               # ...or any account/profile explicitly
janusmcp schema list_tables           # full input schema of ONE tool, only when needed
janusmcp call list_tables --args '{"schemas":["public"]}'
janusmcp call ping --account azienda_b          # cross-account without switching
echo '{"sql":"select 1"}' | janusmcp call db_query   # JSON args via stdin too
janusmcp call ping --json --timeout 30s         # stable envelope for agents

call prints the tool's text content to stdout and exits non-zero on a tool error, so it composes with pipes and scripts. Selectors resolve exactly like in the broker: the persisted active account by default, or any account id / profile name; a name that collides across a profile's accounts must be disambiguated with --account. Secrets resolve through the same vault/OAuth stack as serve — nothing extra to configure.

This makes JanusMCP a first-class citizen for CLI-driven agents (Claude Code, Codex, Cursor agents, or any agent with shell access): instead of registering it as an MCP server, just tell the agent that janusmcp tools / schema / call exist. Discovery, schemas and invocation happen on demand, credentials stay in the keychain, and multi-account switching works the same as over MCP. Both modes share the config and the persisted active account, so you can mix them freely.

For repeated calls, janusmcp daemon start launches an authenticated loopback-only broker. tools, schema, and call auto-detect it and reuse its upstream sessions; --direct bypasses it and --daemon requires it. Set JANUS_DAEMON=auto|require|off to choose a default policy.

Key features

How it's different

The MCP gateway space (MetaMCP, mcp-proxy, IBM ContextForge, …) aggregates different servers behind one endpoint. JanusMCP solves the orthogonal, under-served problem: many identities for the same service, without saturating the model's context, from any LLM, fully local. It's a credential-aware broker, not a flat aggregator.

Built-in connectors

Scaffold an account from a ready-made template with janusmcp add <template> (run janusmcp catalog for the full, up-to-date list). Most use browser OAuth (dynamic client registration); a few need extra setup, noted below.

• Remote, browser login (OAuth): supabase, github, notion, sentry, stripe, hubspot, paypal, linear, vercel, canva, neon, netlify, zapier.
• SSE transport (legacy): asana, monday, intercom, webflow, wix, square, globalping.
• Cloudflare (Streamable HTTP): cloudflare-bindings, cloudflare-observability, cloudflare-radar, cloudflare-builds, cloudflare-browser.
• Google Workspace (bring-your-own OAuth client): gmail, google-drive, google-calendar, google-chat — see go/docs/google-workspace.md.
• Per-account URL: activecampaign (paste your Remote MCP URL).
• Figma: figma-desktop (local, recommended) and figma (remote, restricted — see below).
• Generic building blocks: http-oauth, sse-oauth, supabase-pat, stdio.

Missing one? Add any remote server with http-oauth / sse-oauth, any local one with stdio, or define your own template in ~/.config/janusmcp/templates.json.

Provider notes

ActiveCampaign

ActiveCampaign ships an official remote MCP server with a unique URL per account (ActiveCampaign → Settings → Developer → Remote MCP URL) and browser-based OAuth — a natural fit for the multi-account broker. Add one account per client:

janusmcp add activecampaign ac_clientA   # then paste that client's Remote MCP URL in config.json
janusmcp connect ac_clientA              # browser login

Because the URL is per-account, the template seeds a REPLACE_ACTIVECAMPAIGN_MCP_URL placeholder you must replace with your own URL. Login uses dynamic client registration, so no client ID/secret is needed.

Google Workspace (Gmail / Drive / Calendar / Chat)

Google offers remote MCP servers for Gmail, Drive, Calendar and Chat, but — unlike most providers here — they require your own OAuth client (created in the Google Cloud Console); they don't support dynamic client registration. JanusMCP supports this via the oauthClientId / oauthClientSecret / scopes fields, preset by the gmail, google-drive, google-calendar and google-chat templates:

export GOOGLE_OAUTH_CLIENT_ID=... GOOGLE_OAUTH_CLIENT_SECRET=...
janusmcp add gmail gmail_clientA
janusmcp connect gmail_clientA           # browser login as client A

Full one-time Google Cloud setup (enable MCP APIs, consent screen, Desktop OAuth client) and scope details are in go/docs/google-workspace.md.

Figma

Figma offers two MCP servers, handled differently here:

• Local Dev Mode server (recommended). Figma's desktop app hosts an MCP server on http://127.0.0.1:3845/mcp. It's local, needs no OAuth, and works out of the box: enable it in the desktop app (Dev Mode → Inspect → Enable desktop MCP server) and add it with janusmcp add figma-desktop figma_work. Requires a Dev/Full seat on a paid Figma plan.

• Remote server (https://mcp.figma.com/mcp) — restricted. Figma allowlists the OAuth client_name during dynamic client registration and returns 403 Forbidden to any client that isn't in its MCP catalog (VS Code, Cursor, Claude Code, …). JanusMCP is not (yet) an approved client.

  ⚠️ Workaround — opt-in, use at your own risk. You can make JanusMCP register under an approved name by setting "clientName": "Claude Code" on a remote figma account in your config.json. This impersonates an approved client and may violate Figma's Terms of Service; it can also break whenever Figma updates its allowlist. It is not enabled by default. Prefer the local Dev Mode server above for real work.

  The proper long-term fix is for JanusMCP to be submitted to and approved for Figma's MCP catalog, so no client_name override is needed. This is planned.

Status & roadmap

Alpha — the core is implemented and tested in Go.

• Active-account model, per-session / global / locked scoping
• stdio + Streamable HTTP transports
• MCP 2026-07-28 (stateless core) on both transports, legacy revisions unchanged
• OS-keychain vault + encrypted-file fallback
• OAuth loopback (PKCE) with auto-refresh and per-spawn token resolution
• One-command client install (janusmcp install …), Claude Desktop .mcpb, registry server.json
• Multi-server "profiles" per client (Supabase + GitHub + Slack of Client A at once)
• with_account one-shot cross-account calls
• CLI / code-execution mode (janusmcp tools / schema / call) — invoke tools on demand instead of loading all definitions, to cut token usage
• SSE upstream transport (in addition to Streamable HTTP + stdio) for SSE-only servers
• Pre-registered OAuth clients (bring-your-own client for servers without dynamic client registration, e.g. Google Workspace)
• Built-in connector catalog (25+ services) via janusmcp catalog / add
• Signed, per-OS release binaries & registry auto-publish in CI

See design-broker-mcp-multi-account.md for the full design.

Repository layout

• go/ — the broker (Go). This is the real implementation. Build & docs →
• spike/ — the original TypeScript spike, kept as a verified reference of behavior.
• design-broker-mcp-multi-account.md — architecture & rationale.
• docs/ — current architecture, compatibility contract, testing, ADRs, and operations.

Project documentation

• Documentation index
• Architecture
• Compatibility contract
• Architecture Decision Records
• Testing
• Changelog
• Contributing and security policy

Privacy Policy

JanusMCP runs entirely on your machine. It has no backend servers, collects no data, and contains no analytics or telemetry — the developers receive nothing about you or your usage. Credentials and tokens are stored in your OS keychain (or, if you opt in, an encrypted local file) and are used only to authenticate to the services you configure; they never pass through the model context. Network connections are made only to the MCP servers and OAuth providers you configure. Because everything is local, data is retained only on your own device for as long as you keep it, and removing it (janusmcp vault delete <name>, or uninstalling) removes it entirely.

Full policy: https://janusmcp.dev/privacy.

Contributing

Contributions are very welcome — see CONTRIBUTING.md. New connector presets, client integration guides, and packaging help are especially appreciated.

License

MIT — see LICENSE.