​

MCP Server & Claude Skill

​

What is the MCP Server?

“Create a collection called orders, add these three documents to it, and set up a cron job that runs the send-daily-digest function every morning at 9 AM.”

​

Setup

​

Step 1 — Install globally

npm install -g cocobase-mcp

npx cocobase-mcp

​

Step 2 — Get your credentials

• API Key — project settings → API Key
• Project ID — project settings → Project ID

​

Step 3 — Configure your AI tool

{
  "mcpServers": {
    "cocobase": {
      "command": "npx",
      "args": ["-y", "cocobase-mcp"],
      "env": {
        "COCOBASE_API_KEY": "your-api-key",
        "COCOBASE_PROJECT_ID": "your-project-id"
      }
    }
  }
}

{
  "mcpServers": {
    "cocobase": {
      "command": "npx",
      "args": ["-y", "cocobase-mcp"],
      "env": {
        "COCOBASE_API_KEY": "your-api-key",
        "COCOBASE_PROJECT_ID": "your-project-id"
      }
    }
  }
}

{
  "cocobase": {
    "command": "npx",
    "args": ["-y", "cocobase-mcp"],
    "env": {
      "COCOBASE_API_KEY": "your-api-key",
      "COCOBASE_PROJECT_ID": "your-project-id"
    }
  }
}

​

Environment variables

​

Available Tools

​

Project

​

Collections & Documents

​

App Users

​

Cloud Functions

​

Cron Jobs

​

Storage

​

Email

​

Analytics

​

Migrations

​

Example Conversations

“Create a posts collection, a comments collection, and add these 5 sample posts to get me started.”

“Show me the code for my send-weekly-report function. It’s not sending emails correctly — the template name is wrong. Fix it and test it with payload {'week': '2025-W21'}.”

“Give me all users who signed up in the last 7 days and export them as CSV.”

“Create a cron job that runs cleanup-expired-sessions every night at 2 AM UTC.”

“Show me analytics for the last 30 days — how many new users, what’s my storage usage, and how many function calls did I make?”

​

Claude Skill — Cloud Functions AI

​

How to use the Claude Skill

/home/patrick/Documents/COCOBASE/cocobase-mcp/CLOUD_FUNCTIONS_SYSTEM_PROMPT.md

​

What the Claude Skill knows

• Every global variable available in functions (db, asyncdb, req, email, http, auth, render, queue, use)
• Every db method with correct signatures: get_documents, find_one, create_document, update_document, update_document_fields, delete_document, count_documents, bulk_create_documents, bulk_delete_documents, query, get_or_create_document
• All filter operators (__gt, __gte, __lt, __lte, __ne, __contains, __startswith, __endswith, __in, __notin, __isnull, [or], [or:group])
• asyncdb — async proxy of all db methods, use with await in async def main()
• Email methods and their exact signatures (send_email, send_welcome_email, send_password_reset_email, send_2fa_code_email, send_password_changed_email)
• http.get/post/put/patch/delete/fetch_all — concurrent external requests
• auth.generate_token / validate_token / refresh_token
• queue.add / call_function / has_capacity
• render.render_html / render_template / render_string
• use(function_name, export_name) for cross-function imports
• Pre-imported stdlib: json, datetime, timedelta, math, re, random, secrets, uuid, base64, hashlib, urllib, collections, itertools, decimal, html, etc.
• Security restrictions: no import, no file system, no os/subprocess, SSRF-blocked HTTP

​

Cloud function structure

# Every function must define main() — sync or async
def main():
    return {"success": True}

async def main():
    user = await asyncdb.find_one("users", email=req.get("email"))
    return {"user": user}

​

Key db methods

# Query — returns {data, total, has_more}
result = db.get_documents("posts", filters={"status": "active"}, limit=20)
posts = result["data"]

# One document — returns dict or None
post = db.find_one("posts", id="abc123")

# Create
doc = db.create_document("posts", {"title": "Hello", "status": "draft"})

# Partial update (safe — preserves other fields)
doc = db.update_document("posts", document_id="abc123", updates={"status": "published"})

# Count
total = db.count_documents("posts", filters={"status": "active"})

# Bulk
result = db.bulk_create_documents("posts", [{"title": "A"}, {"title": "B"}])
result = db.bulk_delete_documents("posts", document_ids=["id1", "id2"])

​

Async version (asyncdb)

async def main():
    posts = await asyncdb.get_documents("posts", filters={"status": "published"})
    doc = await asyncdb.create_document("posts", {"title": req.get("title")})
    return {"created": doc["id"]}

​

Next Steps

• JavaScript SDK
• Flutter SDK
• Cloud Functions guide