Agentic stock intelligence platform with a custom MCP orchestration layer

Equity research coverage does not scale with headcount. Producing a defensible daily view across two dozen equity buckets means repeating the same retrieval, cross-checking and write-up work every night, and an LLM asked to do it in one pass produces fluent output with no traceable basis. Two further problems made a naive pipeline unusable: the model had no memory of what it concluded yesterday, so it restated the same thesis as if it were new, and nothing in the loop could distinguish a grounded signal from a confidently hallucinated one before it reached the dashboard.

Built an agentic platform on GPT-4o with a custom MCP orchestration layer exposing four discrete tools: bucket generation, RAG retrieval, LLM health validation and a database write. The agent selects and sequences them itself rather than following a fixed script, and the validation tool sits in the execution path so a signal is checked before it is ever persisted. Added a ChromaDB memory pipeline on text-embedding-3-small so the agent reasons about portfolio drift across sessions instead of repeating itself. Shipped as a containerised five-service system on GCP Compute Engine, running FastAPI, Next.js, PostgreSQL, Redis and ChromaDB via Docker Compose, with APScheduler handling nightly generation and comparison jobs behind a live dashboard.

- 24 equity buckets covered daily, generated end to end with no human in the loop. - Four-tool MCP orchestration replaced a single-pass prompt, so each signal carries a traceable path through retrieval and validation rather than arriving as free text. - An LLM health validation tool in the write path blocks ungrounded signals before they reach the database. - Cross-session memory over ChromaDB removed repeated theses and let the system reason about portfolio drift over time. - Deployed as five containerised services with scheduled nightly generation and comparison jobs, running unattended behind a live dashboard.