Five minutes

A cited answer about your own repository, inside the tool you already use. Three steps; every snippet runs as written.

Key

In Atlas, open Account → API keys and create a key at the Editor level (it reads and writes). The key is shown once.

Keep it in the environment

export RUNLOG_API_KEY="rlka_..."

Seerlka_… reaches every project you belong to, at your access there, 60 requests a minute, and is revoked on the same page. Calls draw on your plan’s allowance; with none, the first metered call answers 402 (Account → Billing).

Terminal

Install the command, bind a folder to a project, send what is in it.

Three verbs

curl -fsSL https://get.runlogai.com/install.sh | sh
export RUNLOG_RELAY_URL="https://runlog-7613480744.us-central1.run.app"

runlog link --project p7Kd2mQx   # the id is the first segment of the project's URL in Atlas
runlog sync                      # prints the cost and waits for yes
runlog watch                     # keeps sending changes as they happen

SeeBefore anything uploads: Reading 412 files (12.4 MB) will use about $2.10 of the $9.00 you have left to spend this week. Nothing is sent until you answer yes. .env and its kin, keys, ignored paths, files over the ceiling, and symlinks stay home and are named in the output. The binding lives in .runlog/, git-ignored; the key stays in the environment.

Your tool

In the project, open Integrations → AI tools and generate a key. This one (rlk_…) is bound to the project and shown once. Paste one of these, then ask: What does this repository do?

Claude Code

claude mcp add --transport http runlog-memory https://runlog-7613480744.us-central1.run.app/v1/mcp/k/rlk_...

mcp.json (Cursor, Windsurf, VS Code)

{
  "mcpServers": {
    "runlog-memory": {
      "url": "https://runlog-7613480744.us-central1.run.app/v1/mcp/ext",
      "headers": { "Authorization": "Bearer rlk_..." }
    }
  }
}

Connector URL (ChatGPT, claude.ai)

https://runlog-7613480744.us-central1.run.app/v1/mcp/k/rlk_...

SeeThe tool calls memory_query and answers with the files and lines it rested on. What you tell it to remember lands through memory_write.