connect-ai-sample-agents

Production-ready starter projects for building AI agents on top of the CData Connect AI MCP server. Each template ships as a working daily-standup agent that pulls live data from Jira, Slack, and Google Calendar via the CData Query MCP and synthesizes a markdown summary — so you can skip the blank-page problem and start from a codebase that already runs end-to-end.

Available templates

| ID | Framework | Language | LLM providers | |---------------|----------------------------|-------------|----------------------------------------| | adk | Google Agent Development Kit | TypeScript | Gemini | | adk | Google Agent Development Kit | Python | Anthropic / OpenAI (via LiteLLM) | | aisdk | Vercel AI SDK | TypeScript | Anthropic / OpenAI / Gemini | | langchain | LangChain v1 | TypeScript | Anthropic / OpenAI / Gemini | | langchain | LangChain v1 | Python | Anthropic / OpenAI / Gemini |

All templates share the same agent contract: on every run, the agent verifies that Jira, Slack, and Google Calendar connections exist on your CData account, drives the OAuth flow in-band for any that are missing, then produces a markdown standup with Yesterday, Today, and Blockers sections.

Quick start

# Interactive — pick a template and language
npx @cdatasoftware/create-agent

# Or specify both
npx @cdatasoftware/create-agent adk --python
npx @cdatasoftware/create-agent langchain --typescript
npx @cdatasoftware/create-agent aisdk

The CLI copies the chosen template into your current directory, then prints the next steps (install deps, fill in .env, run). Templates that exist in both Python and TypeScript prompt you to choose when no --python / -p or --typescript / -t flag is given.

List everything available:

npx @cdatasoftware/create-agent --list

What's in each template

Each scaffolded project includes:

• agent.py or agent.ts — the agent definition (system prompt, tool wiring, model selection, REPL/web entry point).
• .env.example — the env vars you need to fill in (CDATA_ACCT_EMAIL, CDATA_PAT, plus one LLM provider key).
• README.md — runtime-specific setup, run, and customization notes.
• .gitignore — sensible defaults for the runtime (excludes .env, node_modules/, .venv/, __pycache__/, etc.).

The per-template READMEs go into framework-specific detail — model selection, prompt customization, MCP endpoint overrides, etc.

Prerequisites

To run any scaffolded template you'll need:

1. A CData Connect Cloud account and a Personal Access Token (PAT).
2. One LLM provider API key. Each template lists which providers it supports in the table above.
3. Runtime: Node.js 18+ for TypeScript templates, Python 3.10+ for Python templates.

License

MIT — see LICENSE.