MCP Router
Enables AI clients to manage multiple accounts per provider (e.g., Notion workspaces, GitHub identities) through a single MCP endpoint, with sticky account selection and context switching.
README
MCP Router
One MCP server. Unlimited accounts. (working title — see naming note below)
MCP Router is a local-first, open-source MCP server that sits between an AI client (Claude Desktop, Cursor, …) and your other MCP servers, and solves one problem exceptionally well: multiple accounts per provider — three Notion workspaces, two GitHub identities, several Slack teams — through a single MCP endpoint, with account selection that feels native to the model.
In plain terms: AI tools like Claude can connect to apps such as Notion or Slack, but each connection only knows about one account — and most people have several. Two jobs, a client, a personal life. Instead of installing the same connector three times and drowning in duplicate tools, you install the router once and tell it about all your accounts. To the AI it looks like a single, ordinary connection; behind the scenes the router quietly holds one connection per account and delivers each request to the right one. You choose where things go the way you'd tell a colleague — "use my work-1 stuff for now", "actually, search my personal Notion" — or group whole setups into contexts, so switching from work-1 to personal flips Notion, Slack, and everything else in one move. Every reply says which account it used, so nothing lands in the wrong workspace silently, and adding a new account is just a few lines in a config file — the running router picks it up the moment you save.
Works with any MCP client
The router speaks plain, standards-track MCP on both sides — there is nothing
Claude-specific in the contract (a deliberate constraint; see
ADR-012). The account parameter is
ordinary JSON Schema, the "which account?" prompt is an ordinary text result,
and switch_account/switch_context are ordinary tools. Claude Desktop,
Cursor, Windsurf, Zed, VS Code, or a homegrown agent on any model can all sit
upstream. Two qualifiers: the experience scales with the model (reading the
enum, obeying the ask-message, and fanning out across accounts are model
behaviors — weaker models just get asked more often), and each client has its
own ritual for registering a local stdio server.
Design philosophy
Invisible plumbing, visible decisions. The router presents the model with a single coherent identity layer; the machinery stays hidden, but which account a request was routed to — especially a write — is always visible.
Status
v0.2. v0.1 was validated in production (Claude Desktop, three live Notion workspaces — see the ADR-003 validation note). Architecture is recorded in docs/adr — start with ADR-000 and the index.
npm install && npm run build
node dist/cli/index.js init # writes ~/.config/mcp-router/config.yaml
node dist/cli/index.js validate
node dist/cli/index.js run # speaks MCP on stdio
node dist/cli/index.js stats # metrics snapshot of the running router
npm test # 40 tests incl. InMemoryTransport e2e
Working today: config schema + validation, downstream stdio/HTTP clients with
graceful per-account failure, tool merging with the injected account enum,
the five-step route resolver (explicit > sticky > context > singleton > ask),
account markers on implicit routes, switch_account/current_account sticky
defaults, cross-provider contexts (switch_context/current_context/
list_contexts), config hot reload (save the file — changed accounts
reconnect and the tool list updates live; a broken save keeps the last known
good config), auth-aware account health in list_accounts, router stats
(persisted metrics snapshot), router providers|accounts|contexts,
router init --import from claude_desktop_config.json, and
router secret set/rm (OS keychain, hidden input).
Milestones:
- v0.1 (shipped) — multiple accounts, one provider; merged virtual tools;
explicit
accountparameter;router init|validate|run. - v0.2 (shipped) — sticky active account, per-account call health,
router stats. - v0.3 (this) — cross-provider contexts surfaced as tools; config hot reload; expanded CLI.
- Next — Streamable HTTP upstream (per-session state via
Mcp-Session-Id, URL-based client wiring).
Naming: "MCP Router" describes the mechanism, not the concept; a rename is deliberately deferred until pre-v1 (cheap then, a distraction now).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.