Overview

Published on May 25th, 2026

Introduction

The Context Store provides structured context about your software via:

  • A Data Store made up of
    • Integrations — connections to external systems.
    • Data Sources — scheduled jobs to pull data from Integrations and store it as nodes in a graph.
    • Relations — defined rules that link items across systems.
  • Capabilities — runbooks or skills agents can follow.
  • Context Groups — multiple related items collapsed into a single concept.
  • MCP Server Access — the graph, exposed to agents.

Data Store

The Data Store is a centralised database of metadata about the software you create and the authors of that software. Everything from cloud resources from AWS to org charts from Workday.

The Data Store is constructed by Data Sources which use Integrations to retrieve items from a given system of record. The retrieved items are typically structural data such as users, repositories, projects, or accounts are stored as nodes in the context graph.

Start with the Data Store overview and then define how records connect in Relationships.

Capabilities

Capabilities represent a documented procedure that an agent can follow for a well-known process in the organization. Examples include onboarding an employee, rotating credentials, or debugging a failure. You could think of these as Skills or as Runbooks. How they are used and their specific implementation is up to you - the Context Store doesn't prescribe usage.

See Capabilities overview for how to model and organize these runbooks.

Context Groups

Context Groups are a way to collapse multiple items into a single concept. They are used when data in Data Sources doesn't map cleanly or singularly through to what an agent would attempt to retrieve.

For example, an Employee Context Group might combine a GitHub user, an AWS IAM identity, and a BambooHR employee record into one grouped entity.

Read Context Groups overview for recommended grouping patterns.

MCP Server Access

The Context Graph exposes its data through an MCP server. Agents that connect to the server can:

  • Search and navigate the graph
  • Retrieve Context Groups
  • List and follow Capabilities
  • Call Integrations to retrieve live data or perform actions in connected systems

You can configure access in MCP servers overview and then choose the right endpoint for your use case: Explore, Integrations, or Manage.

Next steps