LibreChat Configuration Specialist

LibreChat advanced configuration specialist — AI platform on AWS Overview We are a small startup building a specialized AI assistant platform for a niche B2B vertical. We use LibreChat as our core interface, deployed on AWS, and serve institutional clients who need context-specific AI outputs grounded in their own documents, not generic responses. Think of it as a private, branded AI copilot per client, powered by Anthropic models. We have the foundation running. Now we need someone who knows LibreChat deeply to help us configure Agents, connect MCP integrations, and lock down token management before we go live with our first paying client. What we have AWS EC2 + Docker Compose with LibreChat running in production librechat.yaml v0.1 started (modelSpecs + basic config) Anthropic API live with Claude Sonnet and Haiku models MongoDB + AWS S3 configured Specific tasks for this contract LibreChat Agents configured for RAG via file_search Configure multiple Agents with the file_search tool enabled. Ingest a set of institutional documents as a knowledge base per Agent. Outputs must feel specific to the ingested documents, not generic AI responses. This is a core quality requirement for our product. MCP connectors setup in LibreChat Configure Google Drive MCP (read + draft write), Google Sheets MCP, Google Calendar MCP (read-only), and Gmail MCP (draft-only, never auto-send). Validate that MCP tool calls do not trigger the cache write token bug with Claude Sonnet Agents. Token balance system + per-user quotas Configure LibreChat's native balance/startBalance system per user. Define token limits per Agent. Validate that the known LibreChat cache-write bug with Claude Sonnet Agents is mitigated via max_context_tokens. docker-compose + librechat.yaml hardening Review and stabilize our docker-compose.override.yml and librechat.yaml. Confirm volume mounts are correct. Ensure config changes survive container restarts without data loss. What we need from you You have deployed and configured LibreChat in production before, not just locally. Show us where. You understand the difference between modelSpecs and Agents in LibreChat. If you don't, please don't apply. You are familiar with MCP (Model Context Protocol) and have integrated at least one MCP server. You can work async, document changes in GitHub comments, and are comfortable with a technical lead reviewing every config change before merge. You communicate blockers quickly. We prefer "this is harder than expected, here's why" over 3 days of silence. We are NOT looking for someone to build a new platform from scratch. LibreChat is the platform. Deep config expertise is the job. Known issue we need expertise on There is a known LibreChat bug that causes runaway cache-write token consumption when using Claude Sonnet with Agents that have many tools attached, including MCP connectors. If you have encountered and solved this, please describe it in your proposal. This is our highest priority technical risk. In your proposal, please include A brief example of a LibreChat production deployment you did (link or description). Your experience with LibreChat Agents and file_search specifically. Whether you have worked with MCP servers and which ones. Your take on the cache-write token issue described above.