> ## Documentation Index
> Fetch the complete documentation index at: https://cobo.com/products/agentic-wallet/manual/llms.txt
> Use this file to discover all available pages before exploring further.

# Developer quickstart

> Choose the right integration path for an agent, bot, or automation that needs scoped authority to operate onchain.

CAW in developer mode is simple:

* your runtime gets an API key, not raw private keys
* the wallet owner pairs once in the Cobo Agentic Wallet app
* your runtime submits a pact for a task
* CAW enforces policy rules and approval escalation server-side

That means you can build a bot, agent, or automation that transfers tokens, makes payments, signs messages, or calls contracts without building your own wallet control layer.

## AI coding agent support

If you are building with an AI coding agent (Claude Code, Cursor, Windsurf, etc.), install the CAW developer skill to give it context on the SDK and CLI:

```bash theme={null}
npx skills add CoboGlobal/cobo-agentic-wallet --skill cobo-agentic-wallet-developer --yes --global
```

## Canonical path

If you want the fastest path from zero to a real runnable flow, use this sequence:

1. Install the `caw` CLI
2. Pair with the owner wallet using a pairing code
3. Run the first program in Python or TypeScript
4. Submit a pact, wait for approval, execute a blockchain action, and inspect audit logs

This is the shortest path from zero setup to a real hello-world flow:

<CardGroup cols={2}>
  <Card title="CLI Setup" href="/products/agentic-wallet/manual/developer/cli">
    Download `caw`, pair with the owner wallet, and verify your local setup.
  </Card>

  <Card title="Python Hello World" href="/products/agentic-wallet/manual/developer/api-client-python">
    Run the canonical Python quickstart with pact submission, denial handling, and audit logs.
  </Card>

  <Card title="TypeScript Hello World" href="/products/agentic-wallet/manual/developer/api-client-typescript">
    Run the canonical TypeScript quickstart with the same end-to-end flow.
  </Card>

  <Card title="Pacts" href="/products/agentic-wallet/manual/developer/pacts">
    Understand the authorization object your runtime operates within.
  </Card>
</CardGroup>

## First run outcome

After the canonical path, your runtime should have:

* one paired wallet it can inspect and operate
* one pact approved by the owner
* one successful blockchain action
* one denied action with structured retry guidance
* one audit trail you can inspect programmatically

## Choose your language first

The primary onboarding branch is language, not framework.

<Tabs>
  <Tab title="Python">
    Use this if your runtime logic is Python-first or you expect to adopt a Python agent framework later.

    1. Run [CLI Setup](/products/agentic-wallet/manual/developer/cli)
    2. Run [Python SDK](/products/agentic-wallet/manual/developer/api-client-python)
    3. Expand to [LangChain](/products/agentic-wallet/manual/developer/langchain), [OpenAI Agents SDK](/products/agentic-wallet/manual/developer/openai), [Agno](/products/agentic-wallet/manual/developer/agno), or [CrewAI](/products/agentic-wallet/manual/developer/crewai)
    4. Or use the [MCP Server](/products/agentic-wallet/manual/developer/mcp) if your client already speaks MCP
  </Tab>

  <Tab title="TypeScript">
    Use this if your runtime is Node.js or TypeScript-first, or you plan to integrate with a TypeScript agent framework.

    1. Run [CLI Setup](/products/agentic-wallet/manual/developer/cli)
    2. Run [TypeScript SDK](/products/agentic-wallet/manual/developer/api-client-typescript)
    3. Expand to [OpenAI Agents SDK](/products/agentic-wallet/manual/developer/openai), [LangChain](/products/agentic-wallet/manual/developer/langchain), [Vercel AI SDK](/products/agentic-wallet/manual/developer/vercel-ai-sdk), or [Mastra](/products/agentic-wallet/manual/developer/mastra)
    4. Or use the [MCP Server](/products/agentic-wallet/manual/developer/mcp) if your client already speaks MCP
  </Tab>
</Tabs>

## What you get

Across the canonical flow, CAW gives your runtime:

* scoped onchain authority through pacts
* server-side policy enforcement
* owner approval when thresholds are crossed
* audit logs and structured denial feedback
* MPC signing without exposing key material to your code

## Start with a narrow tool subset

Do not hand your runtime all 20 tools on day one unless it truly needs them. A narrower tool set is easier to prompt, easier to review, and safer to operate.

Start with one of these presets:

* **Pact Drafting** — `submit_pact`, `get_pact`, `list_pacts`
* **Execution** — `transfer_tokens`, `contract_call`, `estimate_transfer_fee`, `estimate_contract_call_fee`, `get_transaction_record_by_request_id`
* **Observer** — `list_wallets`, `get_wallet`, `get_balance`, `list_transaction_records`, `get_audit_logs`

This gives most agentic runtimes a clean split:

* one path to request scoped authority
* one path to perform approved blockchain actions
* one path to inspect state, track results, and explain what happened

## Add a framework only after the hello world works

Once the hello-world path works, choose the surface that fits your runtime:

* **Python SDK** — the primary async path for custom logic and long-running runtimes.
* **TypeScript SDK** — the primary typed Node.js path for backend services and agent runtimes.
* **MCP Server** — best when your client already speaks MCP.
* **LangChain / OpenAI Agents SDK** — good when you want one framework path that can work in Python or TypeScript.
* **Vercel AI SDK / Mastra** — TypeScript-first options for Next.js and Mastra agent runtimes.
* **Agno / CrewAI** — Python-first options for multi-agent and crew-based runtimes.

<CardGroup cols={2}>
  <Card title="Python SDK" href="/products/agentic-wallet/manual/developer/api-client-python">
    Canonical Python hello-world flow.
  </Card>

  <Card title="TypeScript SDK" href="/products/agentic-wallet/manual/developer/api-client-typescript">
    Canonical TypeScript hello-world flow.
  </Card>

  <Card title="MCP Server" href="/products/agentic-wallet/manual/developer/mcp">
    Wallet tools for MCP-compatible clients.
  </Card>

  <Card title="LangChain" href="/products/agentic-wallet/manual/developer/langchain">
    Use CAW with LangChain in Python or TypeScript.
  </Card>

  <Card title="OpenAI Agents" href="/products/agentic-wallet/manual/developer/openai">
    Use CAW with OpenAI Agents SDK in Python or TypeScript.
  </Card>

  <Card title="Vercel AI SDK" href="/products/agentic-wallet/manual/developer/vercel-ai-sdk">
    Use CAW with the Vercel AI SDK in TypeScript.
  </Card>

  <Card title="Mastra" href="/products/agentic-wallet/manual/developer/mastra">
    Use CAW with Mastra agent runtimes in TypeScript.
  </Card>

  <Card title="Agno" href="/products/agentic-wallet/manual/developer/agno">
    Use CAW with Agno in Python.
  </Card>

  <Card title="CrewAI" href="/products/agentic-wallet/manual/developer/crewai">
    Use CAW with CrewAI in Python.
  </Card>
</CardGroup>

## Where advanced modes fit

If you are building an agent, bot, or automation, the path above is all you need. In Agent Mode, your runtime integrates with CAW and Cobo handles the owner-facing pairing, policy, and approval surfaces.

* **Agent Mode** — default path in these docs; available now
* **Core Mode** — for teams that want the core engine without the standard owner-facing product; available upon request
* **Owner Mode** — for teams building their own owner interface; available upon request

Use [Technical Architecture](/products/agentic-wallet/manual/developer/technical-architecture) only when you need to understand those boundaries more deeply.