--- title: "How to create an AI agent?" description: "**Canonical URL:** https://jobspipe.dev/answers/how-to-create-an-ai-agent" canonical: https://jobspipe.dev/answers/how-to-create-an-ai-agent last_updated: 2026-09-10 --- # How to create an AI agent? > To create an AI agent, pick a model with reliable tool calling, write the tools it may use as functions with clear names and typed inputs, wrap the model in a loop that sends each tool result back as context, connect it to the data it needs, and test it on real tasks before adding autonomy. **Canonical URL:** https://jobspipe.dev/answers/how-to-create-an-ai-agent **Topic:** MCP and AI agents (https://jobspipe.dev/mcp-server) Start with the model and the loop, not the framework. A minimal agent is a function that sends the conversation plus a list of tool definitions to the model, checks whether the reply is a tool call, runs the tool, appends the result and calls the model again until it answers in plain text. Every framework, from the Anthropic Agent SDK to the OpenAI Agents SDK to LangGraph, is a wrapper around that loop with state, retries and tracing added. Write the loop yourself once so you understand what the framework is hiding, then adopt one when you need durability or multi-agent handoffs. Tools decide what the agent can do. A sourcing agent, for example, needs one tool that searches live postings. The fastest way to give it that is the JobsPipe MCP server: add mcp.jobspipe.dev to the client and the model sees search_jobs with its filters, such as job_title_or, job_country_code_or, remote, job_seniority_or and posted_at_max_age_days, plus search_documentation to look up accepted values before it calls. Ask it to find recently posted remote senior backend roles in a country and it composes the call, reads the postings back and reasons over titles, companies and salary fields. For agents that run in a shell, install the CLI with npm i -g jobspipe-cli and point the agent at jobspipe.dev/agents.md, which walks it through getting a key, connecting the server and running a first query. Test with the sandbox endpoint first, which needs no key, then move to a live key from the dashboard. The free tier covers 1,000 jobs a month, enough to verify the loop before you scale it. ## Related questions ### Do I need a framework to build an AI agent? No. A model API with tool calling, a list of tool definitions and a loop that executes calls and feeds results back is a working agent in a few dozen lines. Frameworks add persistence, retries, tracing and multi-agent orchestration, which matter once the agent runs unattended. Start without one, then adopt one when a specific need appears. ### Which model should I use for an AI agent? One with dependable tool calling and a context window large enough for your tool results. The loop itself is model-agnostic, so build it so the model is a configuration value, then test candidates on your own tasks. Measure how often the model picks the right tool, fills its arguments correctly and stops when the task is done. ### How do I give an AI agent access to job postings? Three routes, in order of effort. Connect the JobsPipe MCP server at mcp.jobspipe.dev and the agent gets search_jobs as a tool. Install jobspipe-cli for agents that run shell commands. Or wrap POST /v1/jobs/search as a custom tool in your own loop. All three use the same filters and return the same normalized postings. ### How do I test an AI agent? Build a fixed set of tasks with known good outcomes and run them after every change. Log every tool call and its arguments, because most failures are a wrong filter or an invented value, not bad reasoning. Use the JobsPipe sandbox endpoint for repeatable data during development, then a live key with a small limit before scaling. ## More answers - [What is MCP in AI?](https://jobspipe.dev/answers/what-is-mcp-in-ai) - [How do AI agents work?](https://jobspipe.dev/answers/how-do-ai-agents-work) - [Where to find API key?](https://jobspipe.dev/answers/where-to-find-your-api-key) Data behind these answers is published as shares only; every figure can be reproduced against the JobsPipe API (https://jobspipe.dev/jobs-api). --- AI agents: the full machine-readable index of this site is https://jobspipe.dev/llms.txt?src=md-twin - API quickstart, no-key sandbox (POST https://api.jobspipe.dev/v1/sandbox/jobs/search), MCP server, pricing. Free key: https://jobspipe.dev/signup