memeri

Connecting Claude Code

Status: ✅ Current · Last reviewed: 2026-06-22

Claude Code connects in three independent layers, each useful on its own: MCP (required — this alone connects you), user-level telemetry (recommended — surfaces your local file edits in the Control Room), and the plugin (slash commands + cognition / auto-grounding, currently in private beta).

You only need Step 1 to be connected. Steps 2 and 3 are optional add-ons.

What you'll need

  • Claude Code installed (claude on your PATH)
  • Your Memeri token (Connect page → copy)
  • ~2 minutes

Step 1 — Add the MCP server (required)

Run this once in your terminal:

claude mcp add --scope user --transport http memeri {{GATEWAY_URL}}/mcp \
  --header "Authorization: Bearer <your-cb_-token>"

Replace <your-cb_-token> with the token from the Connect page. This registers Memeri with Claude Code at the user level, so it works in every project. Claude Code stores the config itself (in ~/.claude.json).

Don't hand-edit a mcp_config.json file. Current Claude Code doesn't read ~/.config/claude-code/mcp_config.json — use the claude mcp add command above and let Claude Code manage its own config.

Verify: run /mcp inside Claude Code — memeri should be listed and connected. After it makes any tool call, the Connect page Telemetry Health shows MCP = Live.

Step 2 — User-level telemetry (recommended)

This surfaces your local file edits, bash commands, and git commits in the Control Room (without it, you only see MCP tool calls). From your terminal:

mkdir -p ~/.claude
cat > ~/.claude/memeri.json <<EOF
{
  "auth_token": "<your-cb_-token>",
  "gateway_url": "{{GATEWAY_URL}}"
}
EOF
chmod 600 ~/.claude/memeri.json

Every Claude Code window on your machine then forwards activity. Per-project opt-out: touch .claude/memeri-disabled in any folder you don't want tracked.

Step 3 — The plugin (private beta — skip for now)

The Memeri plugin (slash commands + auto-grounding injection) is in private beta and isn't publicly distributed yet — there's nothing to install today, and you don't need it: Steps 1–2 give you full read/write and activity surfacing. When it opens up, the install command will appear here and on your Connect page.

Step 4 — Bind to a job (optional but recommended)

The first time you start work on something:

ask Claude: bind to the job "<job name>"

From then on, every commit, file edit, decision, and update attaches to that job — so your timeline tells a coherent story instead of a flat firehose.

What can go wrong

  • /mcp doesn't list memeri → re-run the claude mcp add command from Step 1 and check the token is current. (There's no mcp_config.json to look for — Claude Code stores MCP config itself.)
  • MCP = Off on Connect → token wrong or revoked; regenerate it on the Connect page and re-run Step 1.
  • Hook = Off but MCP = Live → ~/.claude/memeri.json missing or malformed (Step 2).

For deeper troubleshooting see Telemetry → Troubleshooting.

After you're set up

  • Working With Agents → Posting Good Job Updates — make your AI's progress legible
  • Using Memeri → The Terminal — run agents right in the platform; The Project page — the live view
  • When the plugin opens up, it adds slash commands like /memeri:resume and /memeri:context