Connect AI documentation · Full API reference & configuration details

Scaffold your first AI agent in 10 minutes

Run one command and you have a working AI agent (LangChain, Google ADK, or Vercel AI SDK) wired to Connect AI’s MCP server. The scaffold ships as a daily-standup agent that pulls from Jira, Slack, and Google Calendar and produces a markdown summary. Customize from there.

Prerequisites

• Node.js 18+ (for the TypeScript templates) or Python 3.10+ (for the Python templates)
• A Connect AI Developer Edition account with a Personal Access Token. See the 5-minute guide for the PAT walkthrough
• One LLM provider API key: Anthropic, OpenAI, or Google AI Studio (Gemini)

---

Step 1: Generate the project

Pick a framework and language. Run interactively:

npx @cdatasoftware/create-agent

Or skip the prompts:

npx @cdatasoftware/create-agent langchain --typescript
npx @cdatasoftware/create-agent langchain --python
npx @cdatasoftware/create-agent adk --python
npx @cdatasoftware/create-agent aisdk

Available templates:

Template	Framework	Languages	LLM providers
langchain	LangChain v1	TypeScript, Python	Anthropic, OpenAI, Gemini
adk	Google Agent Development Kit	TypeScript, Python	Gemini (TS); Anthropic / OpenAI via LiteLLM (Python)
aisdk	Vercel AI SDK	TypeScript	Anthropic, OpenAI, Gemini

The CLI copies the chosen template into the current directory and prints next steps.

Step 2: Fill in .env

Each scaffold includes a .env.example. Copy it to .env and fill in:

[email protected]
CDATA_PAT=
ANTHROPIC_API_KEY=sk-ant-...     # or OPENAI_API_KEY, or GOOGLE_API_KEY

Step 3: Install dependencies

For TypeScript templates:

npm install

For Python templates:

pip install -r requirements.txt

Step 4: Run the agent

TypeScript:

npm start

Python:

python agent.py

On first run, the agent verifies your Connect AI account has Jira, Slack, and Google Calendar connections. If any are missing, it drives the OAuth flow in-band. A browser window opens, you sign in, and the agent picks up where it left off. After OAuth completes, it produces a markdown standup with Yesterday, Today, and Blockers sections.

What you got

The scaffold drops a small project into your working directory:

• agent.py / agent.ts: agent definition: system prompt, MCP tool wiring, model selection, run entry point
• .env.example: required env vars
• README.md: runtime-specific run and customize notes
• .gitignore: sensible defaults (excludes .env, node_modules/, .venv/, etc.)

The agent talks to Connect AI via MCP: both the Query MCP (querying Jira/Slack/Calendar data) and the Management MCP (creating missing connections in-band). That’s the same architecture you’d build by hand, just pre-wired.

Customize it

The standup agent is a starting point. Common modifications:

• Swap data sources: edit the agent’s system prompt and connection checks to use Salesforce, HubSpot, NetSuite, Workday, or any of the hundreds of other Connect AI sources instead of Jira/Slack/Calendar
• Change the output: replace the standup synthesis prompt with whatever your agent should produce (a daily revenue report, a customer health summary, a lead scoring run)
• Switch the LLM. Each template supports multiple providers; flip the env var and the model name
• Add tools: extend the agent with non-CData tools (HTTP calls, calculators, file I/O) alongside the MCP-backed query tools

What’s next

• View the templates on GitHub: full source, per-template READMEs, and contribution guide
• By source: source-specific quirks and example queries for Salesforce, HubSpot, Workday, NetSuite, and more
• By AI surface: how Connect AI fits with Claude Code, Cursor, Copilot Studio, and other agent runtimes
• MCP reference: full tool list for the Data and Management MCP servers