AI Automation Developer (Claude Agent SDK, MCP, n8n): build 3 automations and teach me

I'm a freelance data engineer. I want three working automations built and handed over with full source code and setup notes, so I can run and maintain them myself. Please treat this partly as a teaching job: clear documentation and a recorded walkthrough matter as much as the build. THE 3 AUTOMATIONS 1. Pipeline failure investigator (n8n + Claude Agent SDK + MCP) - Trigger: n8n picks up a failed data pipeline run, sent by webhook from a scheduler or dbt. - Investigate: n8n calls an agent built with the Claude Agent SDK. The agent works out the root cause through MCP tools: it reads the error log, queries the warehouse (read-only), and compares row counts and schemas against the last successful run. - Propose: the agent writes a diagnosis and a suggested SQL fix, then opens a draft GitHub pull request with that fix. - Approve: n8n posts the diagnosis to Slack with Approve and Reject buttons. Nothing is merged or re-run without a person approving it. - Must handle: a step limit on the agent's tool calls, retries and timeouts, and a full audit log of every tool call and query. - Test: a set of 5 seeded failure cases (for example: a schema change, a null spike, a duplicate load, a late file, bad credentials), with the agent's hit rate reported for each. 2. Custom MCP server (Python) - A read-only MCP server that lets Claude query a PostgreSQL database. - It needs a table allow-list, a row limit and logging of every query. - Must run locally with Claude Desktop. 3. Custom chatbot for Slack and email (n8n + Claude Agent SDK + MCP) - Incoming Slack: n8n receives direct messages and @mentions in chosen channels. The bot replies in the same thread. - Incoming email: n8n watches a dedicated Gmail address. The bot writes a reply as a Gmail draft for me to approve. Nothing is sent automatically. - Answers questions from a set of my documents (PDFs and Google Drive), and cites the source document and section for each answer. - Uses the MCP server from automation 2 for questions about the data. - Remembers earlier messages within each Slack thread and each email thread. - Says "I don't know" when the documents don't cover the question, and flags it to me instead of guessing. - Only replies to approved Slack channels and approved email senders. - Test: a set of 20 questions with expected answers, reporting accuracy and citation accuracy, run through both Slack and email. - Runs in Docker. DELIVERABLES FOR EACH AUTOMATION - A Git repo, or an exported n8n workflow JSON - A README covering prerequisites, environment variables, setup steps and how to test - A 20-30 minute recorded walkthrough - One live handover call (all 3 can be covered in one call) REQUIREMENTS - Real commercial experience with the Claude API / Claude Agent SDK, MCP and n8n. Personal or hobby projects only won't be enough. - Python - You write documentation that someone else can actually follow BUDGET AND STRUCTURE - Fixed price: [£X total], split into 3 milestones, one per automation - Each milestone is released once I've followed the README myself and the automation runs TO APPLY - Describe one automation you've built commercially with Claude, MCP or n8n, in depth: the problem, the architecture, what broke, and how you fixed it. - Start your proposal with the word LANTERN. Applications without it won't be considered.