Connect an AI assistant

Let Claude or another AI assistant read your guides. Sign in from the assistant, or use a token; disconnect at any time.

Add Nehemvix to Claude

Claude (claude.ai, the Claude desktop and mobile apps, and Claude Code) connects by signing in; no token to copy.

1. In Claude, open Customize → Connectors (Settings → Connectors in some versions) and choose Add custom connector.
2. Name it Nehemvix and enter the address https://nehemvix.aelsha.com/mcp. Leave the advanced OAuth fields empty.
3. Choose Connect. Nehemvix opens: sign in with Google if you aren't already.
4. Check the consent screen: it names the app asking, shows where your answer is sent (claude.ai for the Claude apps) and lists what the access allows. Choose Allow.
5. Back in Claude, turn the connector on in a chat and ask, for example, "list my Nehemvix guides".

Claude Code: claude mcp add --transport http nehemvix https://nehemvix.aelsha.com/mcp, then run /mcp and choose Nehemvix to sign in. Claude Code receives the answer on your own computer (localhost), and the consent screen says so: allow it only if you started the sign-in yourself.

Other assistants that support MCP sign-in (OAuth 2.1 with PKCE, client ID metadata documents or dynamic registration) connect the same way, as long as they receive the answer on localhost or on Claude's address. For anything else, use a token below.

Connected apps

Settings → Connected apps lists every assistant you connected by signing in: its name, the address it answers to, when you connected it and when it last read your guides. Disconnect ends its access at once. A connection that goes unused for 30 days stops by itself; the assistant then asks you to sign in again. Erasing your data removes every connection.

Create a token

For assistants that don't support signing in (or for scripts), use a personal access token.

1. In the app, open Settings → Connect an AI assistant (shown when the service has it switched on).
2. Name the token and choose how long it lasts: 7, 30 or 90 days. You can have up to 10 active tokens.
3. Copy it now: it is shown once and never stored in readable form.
4. MCP: copy the ready-made client configuration shown there (endpoint https://nehemvix.aelsha.com/mcp).
5. A2A: the agent card is at /.well-known/agent-card.json; send requests with the token as a Bearer header.

Example: an MCP client configuration

Settings shows this with your token filled in. Clients that read an mcpServers file (Claude Code's .mcp.json, Cursor and others) take it as it is:

{
  "mcpServers": {
    "nehemvix": {
      "type": "http",
      "url": "https://nehemvix.aelsha.com/mcp",
      "headers": { "Authorization": "Bearer rlp_…" }
    }
  }
}

Keep the file out of version control: the token reads all your guides until it expires or you revoke it. Then ask your assistant, for example, "list my Nehemvix guides" or "search my itsdangerous guide for how tokens expire".

What an assistant can do

• list your guides and their status;
• read a section, with citations to the exact lines;
• search a guide;
• see the readiness checklist;
• prepare a link that opens a new guide with the repository filled in.

It can't start a guide, spend money or see anyone else's guides: you still choose the model, agree to what is sent and approve the cost in the app. Only assistants that call from a server or desktop app work; browser-based clients are blocked on purpose. Each call appears in that guide's How it works, live and, afterwards, under How it was made.

Revoke

Settings → Connect an AI assistant → Revoke next to the token, or Settings → Connected apps → Disconnect for an assistant you connected by signing in. Either stops working at once.

Hosted or local

This page is about the hosted server. Nehemvix also runs as a local MCP server for agents you build:

Details and seven tested frameworks: Use Nehemvix from AI agents. A recorded run of one, with the cost approved before anything is spent: Nehemvix for AI agents.

How the MCP server and the A2A endpoint are built, and how they were tested: MCP vs A2A.