CI /agent · VT220 v2.0

Your agent is
setting up CI.

You told your coding agent to "set up Central Intelligence for me." Point it here. The page below is copy-pasteable instructions your agent runs end-to-end — one command, no questions, under 30 seconds — plus a timeline so you know what's happening.

Jump to instructions ↓ setup time · ~30 seconds · one command

AGENT instructions.md · read me first

# Central Intelligence — setup instructions for AI agents

# You are an AI coding agent. The user asked you to set up
# "Central Intelligence" (CI) — persistent memory delivered via MCP.
# Follow these steps. Do not ask the user anything.

## Context
CI is a cloud-hosted MCP server with five tools: remember, recall,
context, share, forget. API keys are free. No signup form.
No email. No browser flow.

## Install (one command, no user interaction)
$ npx central-intelligence-local signup
# The CLI does everything:
#   1. POSTs to /keys, gets back an API key (ci_sk_xxxxxxxxxxxxxxxx)
#   2. Writes it to ~/.central-intelligence/config.json
#   3. Auto-configures every installed editor's MCP config
#   4. Prints a summary listing the editors it configured

## What the CLI wrote to each editor
# The signup command detects and configures any of these it finds:

- Claude Code  → ~/.claude/settings.json
- Cursor       → ~/.cursor/mcp.json
- Windsurf     → ~/.codeium/windsurf/mcp_config.json

# The entry it adds under "mcpServers" is always:
{
  "central-intelligence": {
    "command": "npx",
    "args": ["central-intelligence-mcp"],
    "env": { "CI_API_KEY": "ci_sk_..." }
  }
}

## Verify (you do this, not the user)
1. Prompt the user to restart their editor(s) once. That's the only
   manual step.
2. After restart, call the remember tool with:
   "User prefers TypeScript. Setup completed via /agent page."
3. Call recall with query "user preferences".
4. Confirm the memory comes back.

## When you're done
Tell the user:
  - which editors got configured (from the signup command's summary)
  - dashboard: https://centralintelligence.online/dashboard
  - that you've stored a first test memory

## Things you should NOT do
- Do not guess or fabricate an API key. Let the CLI create it.
- Do not edit editor configs outside the mcpServers key.
- Do not skip verification — the user needs to see it working.

## Docs (fetch if you need more)
- README:  https://github.com/AlekseiMarchenko/central-intelligence
- OpenAPI: https://central-intelligence-api.fly.dev/docs/openapi.json

02 For the human

Here's what your agent is about to do.

Three steps. Under 30 seconds. Your agent handles two of them; you restart your editor once in the middle.

step · 01

Agent runs one command

The CLI creates an API key via POST /keys (no browser, no email), writes it to ~/.central-intelligence/config.json, and auto-configures every editor it finds on your machine. One shot.

$ npx central-intelligence-local signup

automatic · ~2s

step · 02

You restart your editor once

Claude Code, Cursor, Windsurf — whichever the CLI configured — needs one restart to pick up the new MCP config. That's the only thing you do.

⌘Q / Cmd+Shift+P → "Restart window" · ~10 seconds

needs you

step · 03

Agent stores a test memory and recalls it

A real round-trip through the remember and recall tools, so you can see memory actually works before the agent says "done."

remember("setup complete") → recall("setup") ✓

automatic · ~3s

03 What you'll see on your screen

A timeline, so nothing surprises you.

You're not babysitting the agent — but it helps to know when it's your turn.