Labsco
MCP SERVER

Calypso RAG

by calypso-so ↗

Hosted multimodal RAG in one npx command — upload PDFs, screenshots, charts and diagrams into buckets, then ask grounded questions with citations.

Vector Stores & RAG Retrieval
Summary
No vector stack to run, and visuals are handled natively.

The comparison the README draws is against local-first RAG stacks that need a Python setup or Docker Compose plus a separate vision pipeline before an agent can ask anything. Here the retrieval layer is hosted and images, charts and PDF tables are indexed the same way as text. Two practical details save time: accepted or queued means durable, not query-ready — wait for indexing before asking about fresh uploads — and this server does not write files to your machine, so pair it with a filesystem server if you want reports saved locally.

What it is

An MCP front end for Calypso, a hosted retrieval layer built on Gemini File Search. It handles text and visual content in the same index — PDFs, screenshots, charts, diagrams — so questions can be asked across the formats company knowledge actually lives in, and answers come back with source references and retrieval metadata.

What you get
  • `calypso-rag-agent` — grounded questions against the knowledge base, with multi-turn chaining and `/new` to reset the thread; an optional `model` argument selects a team RAG variant, and each variant keeps its own conversation chain
  • `calypso-list-buckets` — the buckets available to your API key, with ids, slugs, status, member and source counts and store readiness
  • `calypso-upload-file` — a single file into the durable knowledge store, routed by `bucketIds`, `bucketSlugs` or `bucket`, with optional title, tags and metadata, and an option to wait until indexing is ready
  • `calypso-upload-files-batch` — 1 to 100 files in one request against a shared bucket, with per-item overrides and a required `batchIdempotencyKey` so retries resume rather than duplicate
  • `calypso-get-file` — one file's metadata and indexing status, with a `verify` flag that cross-checks against the provider when a status looks stale
  • `calypso-create-bucket` and `calypso-create-agent` — create a destination first, or bind a new RAG variant to buckets; creating an agent refreshes the catalogue and notifies the client, so the variant is usable without reconnecting
  • Read-only resources including `calypso://buckets`, `calypso://rag-agent-models`, `calypso://workflows` and `calypso://security`, plus prompts for asking, ingesting and resetting
Requirements

A Calypso API key beginning `sk-`, passed as `CALYPSO_API_KEY` or the `--api-key` flag; `CALYPSO_API_BASE_URL` is optional and defaults to `https://api.calypso.so/v1`, only worth overriding for a self-hosted deployment. Node.js 18+ — the server uses Web Fetch API primitives and bootstraps any that a given MCP runtime does not provide. Run it as `npx -y @calypsohq/multimodal-rag-mcp-server`; the package is 2.3.0 in package.json. One thing to get right on uploads: use `filePath` for local installs where the server can read your disk, and `contentBase64` for hosted or remote clients that cannot — a path under a hosted sandbox prefix will not resolve.

Setup effort

One command plus a key — npx -y @calypsohq/multimodal-rag-mcp-server --api-key "sk-your-key-here", then supply credentials