AI Agent Developer to Automate Our Business Workflows with Claude & MCP

About us We're a growing company. Our orders, inventory, invoicing and customer accounts run through our ERP, which has a REST API. Our team spends hours every day on manual data entry that AI agents should be doing. We're looking for a senior AI agent developer to design, build and own this with us over the long term. What our team does manually today • Copies order details from emails and PDF documents into the ERP • Creates and updates orders by hand • Checks stock levels and chases suppliers • Follows up on unpaid invoices one by one • Builds the same reports every week What you'll develop PHASE 1: MCP integration + first AI agent • Build an MCP (Model Context Protocol) integration on top of our ERP API, so Claude and other AI agents can work with our data safely • Develop clear agent tools for finding, creating and updating records • Keep read-only tools separate from tools that change data • Role-based permissions, so each agent only gets the access it needs • Log every tool call: who, what, when • First AI agent: reads incoming emails and PDFs, extracts the details and drafts the order. A team member approves it in one click before it's saved. PHASE 2: more AI agents • Stock alerts with draft reorders • Automatic invoice follow-up emails in our tone of voice • Regular summaries posted to Slack or email PHASE 3: internal AI assistant (chat) • Staff ask questions about orders, customers and invoices and get answers from live data • RAG over our internal documents, and it says "I don't know" instead of guessing How we want it built • A person approves anything that touches money, orders or customer emails • Automated tests (LLM evals) measure accuracy before every release • A running cost estimate before development starts, and spend limits in production • Fast responses and processing • Clean code, small pull requests, clear documentation our team can maintain Tech stack Open to your recommendation. We expect something like Python, FastAPI, the Claude API (OpenAI or Gemini where cheaper), LangGraph or the Claude Agent SDK, PostgreSQL and cloud hosting. Use AI only where it adds value. Plain code is fine everywhere else. You have • Developed and shipped at least one AI agent with tool calling to production, not just a demo • Built MCP integrations or similar tool APIs for LLMs • Strong Python, API development, authentication and permissions • Experience extracting structured data from documents with LLMs • Ownership: you plan the work, explain the trade-offs, and report honestly Engagement • Hourly, long-term, with ongoing development after the first phase • Start with Phase 1, prove it works, then expand To apply, please answer: 1. Describe one AI agent you've developed and put in production. What did it automate, and how did you measure that it worked? 2. How would you stop an AI agent from saving wrong data into a business system? 3. How would you keep the running cost of the AI agents low?