AI Coding Agents

Working with AI Coding Agents

Connect Claude Code, Cursor, ChatGPT and other agents to EpicContext over its MCP server. Nothing to install: the server holds the context, the rules and the skills.

EpicContext connects to AI coding agents through its MCP (Model Context Protocol) server at https://www.epiccontext.com/api/mcp. Your agent reads your product context — brand, personas, design system, architecture, decisions — live from the graph, and writes back with a reason. Nothing runs on your machine and there is no package to install or update.

Claude: connector plus plugin

Connect Epic Context

In Claude: Settings → Connectors → Add custom connector, paste https://www.epiccontext.com/api/mcp and sign in. One connection covers all your projects.

Add the plugin

Settings → Plugins → Add marketplace with epicstories/epiccontext-marketplace (in Claude Code: /plugin marketplace add epicstories/epiccontext-marketplace), then install epiccontext. No key or project id to enter.

Link your repo

Save the .epicontext file from Your setup (avatar menu, step 3) in the root of your repo. It holds only the project id, so you can commit it.

In a linked repo every session starts with the project and its first steps, and the agent fetches a briefing before it changes UI — the first UI edit waits until it has. Repos without the file are left alone. Updates arrive with the plugin.

Cursor, ChatGPT, Codex and other MCP clients

Add the MCP server and sign in:

{
  "mcpServers": {
    "epiccontext": {
      "type": "http",
      "url": "https://www.epiccontext.com/api/mcp"
    }
  }
}

On connect the server tells the agent the five working rules and — for a project-bound connection — which project it works on. To keep the anchor in the repo as well, use Project settings → Copy the anchor for AGENTS.md and paste it at the top of AGENTS.md.

Scripts and CI that cannot sign in can use an API key instead: create one under Project settings → API keys and send it as Authorization: Bearer <key>.

The tools you will use most

Tool	What it does
`get_briefing`	The context a task needs, by task type (UI component, page, content, strategy, research, backend) — call it before you write
`get_block` / `search_blocks`	One block by key or name; search by name, alias or text
`create_block` / `update_block`	Write, with a `reason`; the write lands directly and shows in the changes feed
`link_blocks`	Say how two blocks relate: supports, supersedes, contradicts…
`save_rule` / `log_learning`	Keep a rule or a learning in the graph instead of a local memory file
`list_skills` / `get_skill`	How to work: the method skills, and your organisation's design skill
`check_citations`	Check an answer: is every claim carried by the block it cites?

How it works

Your editor or AI tool
      ↓ (MCP over HTTPS, signed in)
EpicContext MCP server
      ↓
Your project's graph — the single source of truth

The agent reads from the graph, so it always works with the latest context.
Every write carries a reason and appears in the project's changes feed, where one click undoes it.
Locked blocks refuse every writer; that is the one hard gate.

Troubleshooting

The agent does not see the project

Use a project-bound connection, or pass project_id in the tool call. list_projects shows the projects your account can reach.

A write is refused

The block is locked, or the write is missing a recent briefing in a project that enforces it. The error says which; fetch the briefing with get_briefing and try again.

Next Steps

Learn about writing effective context for AI agents
Explore all available block types to enrich your context
Set up the Widget for stakeholder feedback in your application

Last updated: 2026-10-17