Search Results for

      Show / Hide Table of Contents

      Use Claude Code, Cursor or Copilot with Intent Architect

      Keep the coding agent you already use and still get Intent Architect's modelling and hard guard rails. Want to stay in your terminal or editor? Connect Claude Code, Cursor, Copilot or another agent to Intent Architect, and it models and generates your code from there. Want to run tasks side by side in their own worktrees, or build from a spec? Pick the same agent as the AI in Intent Architect's chat. Or set up both, and use whichever suits the task.

      This guide compares the two, then shows how to set up each.

      Compare your agent in Intent Architect's chat with your terminal or editor

      In Intent Architect's chat From your terminal or editor
      Where you prompt Intent Architect's AI chat, with your agent picked as the model Your agent, where you already use it
      Agents Claude Code, Codex, GitHub Copilot CLI, Cursor CLI, Kiro, OpenCode Claude Code, Codex, GitHub Copilot CLI, Cursor, Kiro and OpenCode in one click; other MCP agents, such as Copilot in VS Code, by hand
      Subscription Your own Your own
      Modelling and code generation Yes Yes
      A Git worktree for each task Tick worktree in the Agents Shell Create your own
      Spec-Driven Development Yes No
      Questions and approvals Cards you click in the chat Your agent's own prompts
      Build, test and run tasks, app preview Yes Your agent runs its own commands
      Watching and reviewing Agents Shell, Source Control, Change Review The same, with the session shown as an MCP row in the Agents Shell

      Either way, the agent updates the model, runs the Software Factory, then writes the code that isn't generated. If you live in an editor, connect from there. If you want several agents working at once, or a feature taken from requirements to tested code, use the chat. The two don't get in each other's way, so you can set up both.

      Pick your CLI agent as the AI in Intent Architect's chat

      1. Install the agent and sign in to it from a terminal. CLI Agents as AI Providers has the commands for each one.
      2. Restart Intent Architect, which looks for installed agents when it starts.
      3. In the AI chat, open the model picker below the message box and choose a model under your agent's name, such as Claude Code.

      The model picker filtered to Claude Code The Claude Code group in the model picker.

      The first chat with Claude Code, Codex or GitHub Copilot CLI installs a connector for it, which can take a minute. The agent runs on your own subscription and still follows your CLAUDE.md or AGENTS.md, and Intent Architect adds its own agents, such as Plan, and its skills. To give a task its own copy of your code, start the chat in the Agents Shell and tick worktree; see Run agents in parallel.

      With worktree ticked in the Agents Shell and the prompt "Add an optional VatNumber to Customer and include it in the customer DTOs.", Claude Code made the change in its own worktree, with all 60 tests passing. Because a spec covers customers, it also updated that spec, had the new wording reviewed first, and linked the new requirement to the code. Then it asked whether to check the spec now:

      A question card in the chat Claude Code asking a question in Intent Architect's chat, answered with a click.

      Connect your terminal or editor agent on the Intent MCP tab

      An MCP server lets an agent call another app's tools. Connect your agent to Intent Architect's, and it can change the model and run the Software Factory from wherever you use it.

      1. In Intent Architect, open AI Configuration, for example by typing /config in the AI chat, and select Intent MCP.
      2. Choose Repo to connect the agent in this Git repository only, or User to connect it everywhere you use it.
      3. Click Connect on your agent's row. The row then shows Connected.

      The Intent MCP tab with Claude Code connected Claude Code connected at Repo scope.

      At Repo scope, Connect writes the agent's MCP settings file, such as .mcp.json for Claude Code, and adds it to .gitignore, because it holds the path to your Intent Architect installation. For an agent without a row, such as GitHub Copilot in VS Code, add the executable shown under Other agent / manual configuration to the agent's MCP settings.

      Then work as usual in that Git repository. The agent finds the Intent Architect solution, loads Intent Architect's instructions, models the change, runs the Software Factory and writes the remaining code with its own file tools. Intent Architect doesn't need to have the solution open, and if it isn't running, the agent's first call starts it in the background.

      Note

      Claude Code asks you to approve a new server from .mcp.json the first time it starts in that folder. Approve intent-architect.

      From a terminal, with "Add an optional Website to Customer and include it in the customer DTOs.", Claude Code added Website to Customer and to the customer DTO and commands in the designers, ran the Software Factory, wrote the handler and test code, and finished with all 56 tests passing, in under 4 minutes. It noted that it had no spec tools, so it couldn't update the spec that covers customers.

      Follow a terminal agent's session in the Agents Shell

      While an agent connected on the Intent MCP tab works, the Agents Shell's Sessions panel lists its session as an MCP row named after the agent, claude-code here, in its Git repository's group. Select the row to see its folder and branch, and the designers it opened as tabs, so you can watch the model change. Its changes appear in Source Control like any others; click the flag button there to review them in Change Review. The row is kept until Intent Architect restarts.

      An MCP session row selected in the Agents Shell The claude-code session, with the Website attribute it added open in the Domain designer.

      Keep Intent Architect's skills up to date in your agent's folders

      Whether your agent runs in Intent Architect's chat or connects from your terminal or editor, it uses Intent Architect's built-in skills and rules, copied into the agent's own folders, such as .claude/skills and .claude/rules for Claude Code. Connect doesn't copy them. To add or update them, open AI Configuration > Skills and Rules, choose your agent and click Update all. To keep them up to date from then on, choose Auto sync on that tab, or click Enable auto sync on the banner a chat shows when they need attention.

      The Skills and Rules tab 45 skills and rules to update for Claude Code.

      Commit the folders, so everyone on your team gets the same skills.

      Related articles

      • CLI Agents as AI Providers - installing each CLI agent, its permission modes and troubleshooting.
      • Intent MCP Server - what the Intent MCP server gives an external agent.
      • AI Configuration - every tab of AI Configuration.
      • Run agents in parallel - running chats in their own worktrees.
      • Vibe coding, Plan Mode or Spec-Driven Development: when to use which - when to prompt, plan or write a spec.
      • Edit this page
      ☀
      ☾
      In this article
      Back to top Copyright © 2017-, Intent Architect Holdings Ltd