Skip to content
AIMake

Connect Codex to AIMake with MCP

Give each Codex session the repository knowledge, decisions and unfinished work your project has already accumulated.

Connection steps

Connect Codex

Start with a working Codex CLI and an AIMake project in an active paid workspace. Connect GitHub in AIMake for repository-backed context. Compare plans.

  1. Add AIMake

    Run these commands in a terminal on the machine running Codex.

    codex mcp add aimake --url https://api.aimake.dev/mcp
codex mcp login aimake

  2. Authorize access

    Complete the browser flow. Choose the AIMake organization, projects and permissions to grant. Keep access tokens out of prompts.

  3. Check the connection

    Run codex mcp list to check configuration. In Codex, ask AIMake to list your projects; a successful response verifies access.

  4. Load the project skill

    Read the public skill, then save its complete contents at the path below in your repository. Start Codex there and ask it to use aimake-mcp.

    .agents/skills/aimake-mcp/SKILL.md
    Read the AIMake skill

Codex stores MCP configuration in its config.toml. AIMake authorization scopes project records; your Codex permissions govern code execution and tools.

One first prompt

Use aimake-mcp. List my AIMake projects and help me choose the one for this repository. Prepare work to recover its decisions and unfinished tasks. Recommend one bounded next step with its evidence requirements. Save the factual outcome with finish_work; wait for my decision before implementing.

What appears in AIMake

AIMake retains the work session and tool activity. Existing wins and tasks keep their state. Once you authorize implementation, ask Codex to attach actual PR and test evidence to the task for review in AIMake.

Troubleshooting

No projects
Check the authorized account, project selection, membership and active workspace subscription. Reauthorize if access changed.
Login incomplete
Rerun codex mcp login aimake and complete the browser flow on the machine running Codex.
Skill missing
Check the repository path and exact SKILL.md filename. Start a fresh session after installation.
Write rejected
Follow the tool error: prepare expired sessions again, authorize needed scopes, or supply the required review/proof. Only a human can provide a waiver.

Further reading