Skip to main content
All connectors
qbash Connectors / Turbopuffer

Turbopuffer MCP serverand Python AI tasks

qbash hosts a Turbopuffer MCP server you can use in Claude, ChatGPT, Codex, Cursor, Grok Bot and Muse. Developers use qbash Tasks to write cloud-hosted Python functions that pair Turbopuffer with models such as Claude and GPT.

Add the qbash MCP

Use Turbopuffer from Claude Code, Codex, Cursor and other MCP clients. Connect Turbopuffer in qbash, and your agent can work with its data and build Tasks that call it.

claude mcp add --transport http qbash https://mcp.qbash.com/mcp

Then run /mcp in Claude Code and sign in with qbash.

  • List namespaces in the region
  • Read a namespace's schema, size and index status
  • Read a namespace's attribute schema
  • Find the nearest rows to a vector (ANN)
  • Rank rows by BM25 full-text relevance
  • Combine vector and BM25 results with reciprocal rank fusion
Read the docs

Write a Turbopuffer task in Python

A qbash Task is a Python function hosted in the cloud. Python is the runtime and defines the process, and built-in inline prompts run on leading models such as Claude, GPT, Gemini and Grok.

Run a Task on a schedule, from an HTTP request or when a connected app sends an event. Your agents and chat tools call it over MCP.

Each run is logged with its inputs and results, and you can turn data logging off for any Task. Every Task has version control, and each run is locked to the version it ran.

Summarize the top full-text matches

Run a BM25 query against one namespace and ask a saved prompt to summarize the rows that came back.

main.py
result = qbash.integrations.turbopuffer.query_full_text(namespace="docs", attribute="title", query="refund policy", top_k=5, include_attributes=["title"])
if result.get("rows"):
summary = qbash.ai.run_prompt(slug="turbopuffer-match-summary", variables={"rows": result["rows"]})
qbash.output(summary)
else:
qbash.output("No matching rows.")
Sample query_full_text response
{
"rows": [
{
"id": 1,
"$dist": 2.1,
"title": "Refund policy"
},
{
"id": 2,
"$dist": 1.4,
"title": "Returns and exchanges"
}
],
"billing": {
"billable_logical_bytes_queried": 1024,
"billable_logical_bytes_returned": 128
},
"performance": {
"server_total_ms": 14
}
}

Custom MCP server

Build a custom MCP server with Turbopuffer

Create a hosted MCP server that combines Turbopuffer tools with your other connectors, Tasks and memory folders. You choose each tool it exposes, so a server can share Turbopuffer reads with write tools left off.

Share the server with your team, and each person adds it to the client they already use.

How to set it up

  1. Connect Turbopuffer on the qbash Connectors page.
  2. Create an MCP server and add the Turbopuffer tools you want to share.
  3. Paste the server URL into your client. It connects with OAuth or an access token.

Turbopuffer MCP questions

What does it cost to use Turbopuffer?

Turbopuffer has no free plan or trial. The Launch plan has a $16 per month minimum usage commitment, and billing is based on storage, writes and queries.

Can a Task delete data in Turbopuffer?

No. The connector can upsert and patch rows. It has no method to delete namespaces or rows, and it cannot branch or copy namespaces.

Which credentials does it need?

An API key from the Turbopuffer dashboard and the region of your namespaces, such as gcp-us-central1.

Build a Turbopuffer task

Write it in Python with AI steps, or describe it to your coding agent through the qbash MCP.