Production, in your AI assistant.

What MCP is

MCP (Model Context Protocol) is the open standard AI assistants use to call tools from other services.

Without it, you open Relia, read the incident, copy the fix plan and paste it into your assistant. With Relia's MCP server, your assistant does that itself: you ask a question in plain words, the assistant decides which Relia tool to call, Relia answers from your project's data, and the assistant uses the answer to explain the problem or edit your code.

1. 1

   You ask your assistant

   “Use Relia: what's broken in production?”

2. 2

   It calls Relia

   relia_status, then relia_get_fix_plan for the incident it found. Relia checks your token and your access to the project on every call.

3. 3

   It fixes it where you code

   “Checkout orders fail since 10:02 for 38 shoppers: service.pyline 88 reads a missing field. Here's the change.” After you deploy, Relia checks live traffic and tells you whether the fix worked.

What your assistant can do

Seven tools, all read-only. Your assistant picks the right one from your question.

1. Create a token

One token per project. It acts as you, can only read, and is shown once.

1. 1

   Open your project in Relia

   Settings, then Advanced, then Integrations. The card Use Relia in your AI assistant is at the top. You can also get there from the Ask Relia page or from any fix plan.

2. 2

   Name it and press Create a token

   For example “Cursor on my laptop”. Relia shows the exact setup line for your tool with the token filled in. Copy it now: it is never shown again.

2. Add it to Claude Code

Run this once in your terminal, inside the project's folder. It stays private to you and to that folder. Don't add --scope project: that writes the token into a file you commit.

claude mcp add --transport http relia https://devreliaapi.tryrelia.com/mcp --header "Authorization: Bearer relia_mcp_…"

2. Or add it to Cursor

Put this in .cursor/mcp.json inside the project's folder, and add that file to .gitignore (it holds the token). Not the global ~/.cursor/mcp.json: a token is for one Relia project.

{
  "mcpServers": {
    "relia": {
      "url": "https://devreliaapi.tryrelia.com/mcp",
      "headers": {
        "Authorization": "Bearer relia_mcp_…"
      }
    }
  }
}

2. Or add it to VS Code

Put this in .vscode/mcp.json inside the project's folder. VS Code asks for the token once and keeps it out of the file, so the file is safe to commit.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "relia-token",
      "description": "Relia token",
      "password": true
    }
  ],
  "servers": {
    "relia": {
      "type": "http",
      "url": "https://devreliaapi.tryrelia.com/mcp",
      "headers": {
        "Authorization": "Bearer ${input:relia-token}"
      }
    }
  }
}

2. Or any other client

Windsurf, Zed, your own agent: anything that supports a remote MCP server with a header.

URL:        https://devreliaapi.tryrelia.com/mcp
Header:     Authorization: Bearer relia_mcp_…
Transport:  Streamable HTTP

Set up Relia from your editor

Haven't installed Relia in this repository yet? Your assistant can do it: say “Set up Relia”.

1. 1

   It asks you two questions first

   Record screen replays? (Relia's agent doesn't need them; recommended no.) Tell Relia who your users are? (Incidents then name customers and plans; recommended yes.)

2. 2

   It shows every change before writing it

   Relia's relia_setupgives it this project's exact steps for your stack (Node, Next.js, web apps, Python). With yes to users, it finds your sign-in code and adds the lines there. Nothing is committed.

3. 3

   It checks the result

   Start your app and use it once; relia_check_setup says what arrived and what is left. In Claude Code it is also a ready-made prompt: /mcp__relia__setup.

Things to ask

Start with “Use Relia” so your assistant knows where to look.

• “Set up Relia.”
• “Use Relia: what needs my attention in production?”
• “What broke in checkout today, and how do I fix it?”
• “Get Relia's fix plan for that incident and apply it.”
• “Is the orders API still failing after my deploy?”
• “Which API is slowest today, and why?”

Safety and limits

• Read-only. No tool can change anything in Relia or on your servers. Your assistant edits your code only in your editor, with your approval.
• One project per token. Another project's incidents are “not found”, even when you can see them in Relia.
• It acts as you. Lose access to the project and the token stops working. Viewers can't create tokens.
• Stored safely. Relia keeps only a hash of the token. Revoke it any time from the same card; it also expires after 90 days without use.
• Every read is logged. Each tool call is an entry in your project's audit log: who, which tool, when.
• Fair use. 300 calls an hour per token. Fix plans and Ask Relia count against your plan's monthly allowance, like in the app.

Questions

Relia's SDKs are open on npm (relia-sdk) and PyPI (relia-sdk). New to Relia? Set it up in five minutes.