add-vibebase

name	add-vibebase
description	Add vibebase — contextual GIF and meme search — as an MCP tool so the agent can respond with reaction media instead of plain text.

Add vibebase

Adds vibebase as an MCP tool. Vibebase is a semantic GIF database — it searches by conversational context (not keywords) and returns the right reaction GIF for the moment.

After installation, the agent gets two tools: search_gif to find GIFs and ingest_category to seed the database with new categories.

Phase 1: Pre-flight

Check if already applied

Look for container/skills/vibebase/SKILL.md in this repo. If it exists, skip to Phase 3.

Get vibebase path

Ask the user:

What is the absolute path to your local vibebase repo? (e.g. /Users/you/code/vibebase)

Store this as <vibebase_path>.

Verify vibebase

1. Confirm <vibebase_path>/mcp.ts exists
2. Confirm <vibebase_path>/node_modules exists (if not, run npm install in that directory)

Verify API keys

Check that the following are set in this repo's .env:

• GIPHY_API_KEY — needed for GIF ingestion from Giphy
• GOOGLE_API_KEY — needed for Gemini vision (describing GIFs) and embeddings (search)

If either is missing, ask the user to add them.

Phase 2: Apply changes

1. Copy the agent-facing skill

mkdir -p container/skills/vibebase
cp <vibebase_path>/nanoclaw/container/vibebase/SKILL.md container/skills/vibebase/SKILL.md

2. Add vibebase to .mcp.json

Add this entry to .mcp.json under "mcpServers":

{
  "vibebase": {
    "type": "stdio",
    "command": "npm",
    "args": ["--prefix", "<vibebase_path>", "run", "mcp"],
    "env": {
      "VIBEBASE_DB_PATH": "<vibebase_path>/gifs.db",
      "GIPHY_API_KEY": "${GIPHY_API_KEY}",
      "GOOGLE_API_KEY": "${GOOGLE_API_KEY}"
    }
  }
}

Replace <vibebase_path> with the actual path.

About VIBEBASE_DB_PATH: Each agent can have its own GIF database by setting a different path. The default uses a single shared database at <vibebase_path>/gifs.db. In a multi-agent setup, you could give each agent its own vibebase.db in its group workspace.

3. Add environment variables

Add to .env:

# Path to vibebase GIF database
VIBEBASE_DB_PATH=<vibebase_path>/gifs.db

Add to .env.example:

VIBEBASE_DB_PATH=path/to/vibebase/gifs.db

Phase 3: Verify

Restart the agent

Restart your agent process. For example:

macOS (launchctl):

launchctl kickstart -k gui/$(id -u)/com.<your-agent>.agent

Linux (systemd):

systemctl --user restart <your-agent>

Test it

Ask the agent:

Use vibebase to find a gif for someone who just shipped their first project.

The agent should call search_gif and return a URL. If the database is empty, it should call ingest_category first, then retry the search.

No git merging needed — vibebase runs entirely as an external process via MCP.