# 60 Second Quick Start > Build your first agent on AgentPMT and start running tools, in two paths — hosted or local. Content type: documentation Source URL: https://www.agentpmt.com/docs/getting-started/quick-start Markdown URL: https://www.agentpmt.com/docs/getting-started/quick-start?format=agent-md Category: Getting Started --- # 60 Second Quick Start You will leave this page with a working agent. Pick the path that matches how you use AI today: - [Path 1 — Hosted Agent](#create-your-first-hosted-agent) - Build and run an agent in your browser. No install. - [Path 2 — Local Agent](#connect-a-local-agent) - Connect your existing Claude, Cursor, or Codex via MCP. Both paths use the same **Bearer Token** from your Agent Guardrail. You can do one now and the other later. --- ## Create Your First Hosted Agent A hosted agent runs inside AgentPMT. You build it once, then use it from the dashboard, the website, the mobile app, or via API. ### Tools and Workflows in 30 seconds **Tool** — one action your agent can take. "Search the web." "Send an email." Pay-per-call. **Workflow** — several tools chained into a reusable recipe. "Research a topic and email me the summary." Runs in one go. Hosted agents pick the right tool or workflow on their own based on your message. ### Build the agent You can start from the marketplace or from the Agent Builder. The fastest path is the marketplace. ### Start from the marketplace 1. **Open the marketplace and click Build custom agent** Go to the [marketplace](/marketplace) and click **Build custom agent** at the top. AgentPMT enters Agent Builder Mode and creates a new draft agent for you. 2. **Add tools and workflows with one click** Stay on the marketplace. Every card now shows an **Add to Agent** button. Click it on anything you want the agent to be able to use. A floating panel keeps your draft's name and composition visible. 3. **Open the agent and try it** From the floating panel, open the [Agent Builder](/agent-workflow-skills-builder?mode=agent) on your draft, give the agent a name and a one-line description, then scroll to the **Scoped chat preview** section and click **Open Chat** to chat with it live. ### Start from the Agent Builder 1. **Open the Agent Builder** Go to the [Agent Builder](/agent-workflow-skills-builder?mode=agent). It opens in Agent mode with a blank canvas. 2. **Name your agent and write a one-line description** The description is what the agent reads to decide how to behave. Be concrete: "Researches market trends and emails a daily digest" beats "research helper." 3. **Attach tools and workflows from the catalog** The canvas has its own **Catalog** with **Available tools** and **Available workflows** sections. Click **Add tool** or **Add workflow** on anything you want. To browse the full marketplace without losing your draft, click **Add tools from marketplace** or **Add workflows from marketplace** — the marketplace opens in Agent Builder Mode and every card shows **Add to Agent**. 4. **Click Open Chat to try the agent** Scroll to the **Scoped chat preview** section and click **Open Chat**. The chat loads against your exact tool and workflow set; ask it to do something that uses one of your attachments. You should see the tool call appear in the conversation and a corresponding entry in **Control Center → Agent Activity**. [Tip title="Start with one tool"] Attach a single free or low-cost tool first, confirm a chat call goes through end to end, then add more. That is the whole flow. See [Using Agents](/docs/getting-started/using-agents) for every place you can run this agent next. --- ## Connect A Local Agent If you already run Claude Code, Cursor, Codex, Zed, or any other MCP-compatible client, you do not need to build a new agent. Connect AgentPMT as an MCP server and the same tools and workflows become available to your local runtime. ### Get your Bearer Token 1. **Open Agent Guardrails** From the dashboard, go to **Control Center → Agent Guardrails**. This is where every Agent Guardrail (your spending and access boundary) lives. 2. **Create a guardrail or pick an existing one** Give it a name that matches the agent ("Cursor", "Local Codex"). Set a spending cap. Add the tools and workflows your agent should be able to call. 3. **Click Show Bearer Token and copy it** The Bearer Token is the only credential your agent needs. Treat it like a password. ### Connect over MCP ### Remote (HTTP) For Claude.ai, ChatGPT custom connectors, and any client that supports streamable HTTP MCP: ``` Endpoint: https://api.agentpmt.com/mcp Auth: Bearer ``` Add it as a remote MCP server in your client. The client will fetch the tool list automatically. ### Local STDIO For Claude Code, Cursor, Codex, and Zed, install the connector once: ```bash npm install -g @agentpmt/mcp-router agentpmt-setup ``` The setup command writes the right MCP config block for your client and prompts for your Bearer Token. Re-run `agentpmt-setup` any time you switch clients or guardrails. ### Make your first call In your agent, ask for a tool you added to the guardrail. For example: *"Use AgentPMT to search the web for `mcp specification` and summarize the top result."* You should see: 1. The agent select a tool from the AgentPMT MCP server 2. A response in your agent 3. A new entry in **Control Center → Agent Activity** within a couple of seconds If you do not see the activity entry, see [Using Workflows and Tools — telling a local agent what to use](/docs/getting-started/using-workflows-and-tools). --- ## Where to next - [Build a Workflow](/docs/getting-started/build-a-workflow) - Chain tools into a reusable recipe. - [Using Agents](/docs/getting-started/using-agents) - Run your agent from the dashboard, the marketplace, mobile, or API.