Skip to main content
When you create a new HubSpot project using the CLI or Developer MCP server, HubSpot scaffolds a set of markdown context files directly into your project. These aren’t code. They’re instructions and reference material written for your AI agent, giving it an informed starting point before it writes a single line of code. For context on why these files matter and how they fit into the broader picture, see Your agent instructions file in the foundations guide. Three files are included because each serves a different role. Two of them (AGENTS.md and CLAUDE.md) are instructions files that agents load automatically at the start of each session. The third (HUBSPOT_PROJECTS.md) is a reference document the agent consults when it needs to understand project structure. Within the instructions files, HubSpot provides both a universal version and a Claude Code-specific version so you can tailor guidance per tool as your project grows. You don’t need to do anything to activate them. Your agent reads them automatically at the start of each session. What you can do is customize them over time as your project develops and your preferences become clear. See Managing your project for how to build on these files as your project grows.

AGENTS.md

AGENTS.md is the universal agent instructions file. Any AI coding agent that follows the AGENTS.md convention reads this file automatically at the start of each session. HubSpot’s version gives the agent platform-level rules it needs to build correctly, including project structure, how to fetch data in a HubSpot project, which CLI commands do what, and what to check before writing anything. This file is your primary way of giving every agent, regardless of tool, the same baseline understanding of your HubSpot project.

CLAUDE.md

CLAUDE.md is Claude Code’s equivalent of AGENTS.md. When both files exist in a project, Claude Code reads CLAUDE.md first. If you’re only working with Claude Code, the two files will look nearly identical to start. That’s expected. The reason both exist is flexibility: as your project grows, you may want to tailor instructions for specific tools. Claude Code supports features that other agents don’t (like slash commands and MCP tool references), so CLAUDE.md gives you a place to add that guidance without cluttering the general AGENTS.md. You might also be working with multiple agents on the same project (a colleague using Cursor, for example), and AGENTS.md ensures they get the same baseline context.

HUBSPOT_PROJECTS.md

HUBSPOT_PROJECTS.md is a reference document for any agent working in the project directory. Unlike AGENTS.md and CLAUDE.md, which are loaded based on tool-specific conventions, HUBSPOT_PROJECTS.md uses a neutral filename that any agent can find when exploring the project. This makes it a useful orientation document for agents that don’t follow those conventions, or for any agent that needs a lighter conceptual overview of the HubSpot project framework without loading all of the technical rules in the other files. Think of it as the orientation document for the project itself, separate from the instructions for the agent doing the work.

Building on these files

The HubSpot-provided files cover the platform side of things: how HubSpot projects are structured, what the CLI commands do, and how to build correctly for the platform. What they don’t cover is your project specifically. As you build, it’s useful to layer your own context files on top of these. A few examples:
  • Spec files: plain-language descriptions of what your project does, what features it has, and how they work. These give fresh agent sessions a high-level picture without needing to re-read the code. See spec-driven development for more on this approach.
  • An issues log: a running record of bugs you’ve hit and how you fixed them. Instructing the agent to check this before investigating any new problem prevents it from re-solving the same issue in different ways across sessions.
  • Plan files: a record of decisions made and features still in progress, so the agent knows what’s done and what’s up next.
You can also customize AGENTS.md for your own working style. For example, if you’re newer to coding and want the agent to explain what it’s doing in plain language as it works, adding that to AGENTS.md means you don’t have to ask every session. See Managing your project for guidance on building these files out over time.
Last modified on September 9, 2026