---
title: "Use Claude Code, Codex or Cursor with Corefinity"
url: https://docs.corefinity.com/ai/coding-agents
type: page
last_updated: 2026-11-13
feature: coding-agents
---
# Use Claude Code, Codex or Cursor with Corefinity

Connect a coding agent to your organisation and to these docs, and give it a project rule to follow.

A coding agent works with two servers: the Corefinity MCP server for your organisation, and the docs MCP server for looking things up.

1. **Connect your organisation**

   Add the Corefinity MCP server to your agent and sign in: see [Connect an AI assistant](/mcp/connect). On the approval page, pin the grant to a staging environment: see [Limit a grant](/mcp/limit-a-grant).

2. **Connect the docs**

   Add the docs MCP server under the name `corefinity-docs`: see [Docs MCP server](/ai/docs-mcp-server). It needs no sign-in.

3. **Add a project rule**

   Put this in `CLAUDE.md` (Claude Code), `AGENTS.md` (Codex) or `.cursor/rules/corefinity.mdc` (Cursor).

   ```markdown
   ## Corefinity

   - This project is hosted on Corefinity. Use the `corefinity` MCP server for the
     organisation and `corefinity-docs` for documentation.
   - Before the first call to a tool, read its guide: `corestack://tools/{name}`.
   - Look up endpoints and tools with `search_docs`. Do not guess names.
   - Work on the Staging environment. Never change production without asking me.
   - Ask me before sending any `confirm_*` value. Never set `reveal_to_agent`.
   - Long work answers a task. Poll the task; do not call the tool again.
   - Never write a token or password into this repository.
   ```

## What to grant

| For                                      | Scopes                  | Pin                     |
| ---------------------------------------- | ----------------------- | ----------------------- |
| Reading logs and settings while you code | `read`                  | The site                |
| Deploying and testing on staging         | `read`, `write`, `exec` | The staging environment |

Before you grant more, see [Safety](/mcp/safety).
