OneTick Blog

Part 2: The MCP Server and Open Access in OneTick 1.27 – Nothing Gets Lost in Translation

Written by Peter Simpson | Nov 20, 2026, 1:45:00 PM

By Peter Simpson, OneTick Product Owner, KX

OneTick 1.27 lets quants and developers describe what they want in natural language and get working OneTick code back, through a new MCP server that plugs into Claude, Codex, GitHub Copilot and other AI agents. Around it, the release widens every path into your data: richer SQL, faster Arrow delivery and easier Python.

In Part 1 we covered speed and storage. This post covers access: how fast your team can go from a question to an answer.

The OneTick MCP Server: An AI Agent That Knows Your Data

The Model Context Protocol (MCP) lets AI assistants call external tools. The OneTick MCP server gives an agent four kinds of knowledge it would otherwise lack:

  • Syntax: how to write a given query in OneTick Python or OneTick SQL.
  • Coverage: which databases, tables and fields exist, and the allowed values of enumerated fields. Including the methodology for derived feature sets.
  • Symbol lookup: which database, and which symbol, holds the instrument you are asking about.

You type a question in VS Code, Claude or another agent, and get back code you can run. The answers rest on OneTick’s public SQL and Python documentation, which holds hundreds of examples, from single functions to full business use cases, with new ones added every week.

There are two ways to use it. The hosted server at https://mcp.cloud.onetick.com/mcp explores the Market Data on Demand catalog, and install instructions for each agent are on onetick.com. To work with your own data, install the onetick-mcp package from PyPI in your environment (Python 3.12 or later) and supply your own coverage, schemas and symbol lookup. 

Built for Regulated Environments

A self-hosted MCP server can keep everything inside your walls. The server generally does not run queries itself: it returns syntax, and you run that code with your own credentials against your own OneTick servers. Documentation and your own examples load into a local PostgreSQL database with the pgvector extension. You can create the embeddings with a Hugging Face model running locally, so nothing leaves your environment, or with an OpenAI model, including OpenAI on Azure, which supports GDPR compliance.

Teams working beyond equities, futures and options can add their own examples, for FX and fixed income, so the agent learns their conventions.

More Expressive SQL

SQL keeps closing the gap with OneTick’s native graph queries. Highlights in 1.27:

  • Time joins with keys. Join trades to the quote from the same exchange, or join a European instrument to quotes or FX rates in its own currency, using the USING syntax.
  • Subqueries in joins. Each side of a join can be plain SQL, an OTQ called from SQL, or a chained query, so any OneTick capability can feed a SQL join.
  • Union All. To allow querying across venue, creating custom consolidated tapes.
  • New functions: GREATEST, LEAST, and GET_SYMBOL_NAME_IN_DB, which shows the symbol as stored, for example the actual contract behind a continuous contract.
  • Exponentially weighted averages can now be specified by number of lookback periods, the approach familiar from pandas and other SQL tools.

Under the hood, the engine that turns SQL into an execution graph has been rewritten and is much faster.

Faster, More Open APIs

REST is now the main delivery path, and it gains the most:

  • SQL over REST. SQL queries previously ran from Python via onetick-py; now they run over REST too.
  • Leaner Arrow responses. REST now sends one Arrow frame per schema instead of one per symbol, a big saving when pulling many symbols with few rows each.
  • Arrow Flight with OAuth 2.0 client-credentials authentication, available in OneTick Cloud.
  • Better drivers and integrations: improved varstring support in ODBC, JDBC and ADBC, query properties exposed in the Trino plugin, and support for Python 3.14.

In Python, pandas DataFrames are now easier to use as data sources. Upload one, join it with market data already on the server, or use it as the symbol list for a query.

Works with the Rest of Your Stack

OneTick 1.27 meets your market data where it already lives:

  • Apache Iceberg. New READ_FROM_ICEBERG and WRITE_TO_ICEBERG event processors read and write external Iceberg tables. External Iceberg tables can be exposed as a view, so it looks like a OneTick archive. Iceberg support will continue to expand in the coming year through 1.28.
  • kdb+. READ_FROM_KDB supports more q data types, and OneTick can write to a kdb-x tickerplant or to Reliable Transport, with better handling of cancellations, corrections and heartbeats. A plugin lets kdb+ users query OneTick through SQL over REST with Arrow transport, the same route used by OneTick Cloud.

Next in the series: the depth features, including order book analytics, real-time collection, entitlements, dashboards and deployment.