Skip to content

Agent integration

Connect Codex to your Agiflow board

Connecting Codex to Agiflow over MCP lets the coding agent read and update Agiflow projects, work units, tasks, and comments, scoped by the connection URL you authorize.

Add to CodexView setup guide

Use Agiflow as the board of record while Codex works through browser-authenticated MCP access from the CLI or IDE.

Codex, connected to your Agiflow board.

Give Codex the board context behind the code

Codex is OpenAI's coding agent for software development, available across the Codex CLI, IDE extension, Codex app, cloud tasks, and ChatGPT. Connect it to Agiflow and Codex can work from the project context your team already maintains: projects, work units, tasks, and task comments.

Agiflow exposes that context through a remote HTTP MCP server with OAuth browser consent. You choose the Agiflow scope before connecting, then Codex acts through the tools available to that authorized connection. Agiflow stays the system of record while Codex uses the board to plan, update, and report coding work.

What Codex can do with Agiflow

  • Inspect Agiflow projects and the work units that break initiatives into shippable pieces.
  • Create, list, update, reorder, and batch-create tasks from a Codex session.
  • Read and write task comments so implementation notes and decisions stay attached to the board.
  • Coordinate workflow locks for long-running or multi-agent work.
  • Use project artifacts where the authorized Agiflow scope allows it.
  • Use vault entries at project scope when the connected user’s role allows vault access.
What Codex cannot do on its own:

Codex can only use the tools and resources exposed by the Agiflow connection you authorize. Agiflow uses resource-bound consent for the selected organization, project, work unit, or task; it does not grant broad named scopes, and role-based controls still apply to vault and project resources.

Connect Agiflow to Codex

Configure the Agiflow MCP server once in config.toml and Codex can use it from both the CLI and the IDE extension.

Which endpoint to use:

For Codex and other coding agents, paste the Agiflow MCP connection URL copied from Settings → Connections rather than guessing a generic endpoint. The coding-agent endpoint is https://agiflow.io/api/v1/mcp. Do not use /api/v1/mcp-app/0.0.1 — that endpoint serves the ChatGPT app/widget MCP surface.

Codex CLI and IDE via config.toml

OpenAI's MCP tutorial documents HTTP MCP servers through config.toml, and the Codex CLI and IDE share the same config file — so this is the primary path.

1

Install or open Codex and sign in

Terminal
curl -fsSL https://chatgpt.com/codex/install.sh | sh

On macOS or Linux, install the Codex CLI with the command above. Authenticate with a ChatGPT account that includes Codex or with an API key.

2

Copy your scoped Agiflow connection URL

In Agiflow, open Settings → Connections and copy the connection URL for the organization, project, work unit, or task you want Codex to use.

3

Add Agiflow to config.toml

~/.codex/config.toml
[mcp_servers.agiflow] url = "YOUR_AGIFLOW_URL"

Add the block to ~/.codex/config.toml, or to a trusted project's .codex/config.toml.

4

Authenticate the MCP server

Terminal
codex mcp login agiflow

Complete the Agiflow browser sign-in and approve the selected resource.

5

Verify the server is available

Terminal
codex mcp list
6

Try a first prompt

Codex
What projects are in my Agiflow workspace?
Using the Codex IDE extension?:

The Codex CLI and IDE extension share config.toml. In the IDE, open the gear menu, choose MCP settings, then Open config.toml. Add the same [mcp_servers.agiflow] block, save, and sign in with codex mcp login agiflow if the server requires OAuth.

Optional CLI shortcut

OpenAI's CLI reference documents codex mcp add --url for streamable HTTP servers, while the MCP tutorial demonstrates the config.toml path. Treat this as version-dependent, not the guaranteed path.

Terminal
codex mcp add agiflow --url YOUR_AGIFLOW_URL codex mcp login agiflow

Recent Codex versions may also support adding a streamable HTTP server from the CLI. If this command is unavailable, use the config.toml setup above.

Troubleshooting

Agiflow is not listed
Check that the table is named [mcp_servers.agiflow], the TOML syntax is valid, and the file is in ~/.codex/config.toml or a trusted project .codex/config.toml. Run codex mcp list.
Browser sign-in loops
Run codex mcp logout agiflow, then run codex mcp login agiflow again and complete the browser flow.
Codex sees the wrong board
Copy a fresh scoped connection URL from Agiflow Settings → Connections for the intended organization, project, work unit, or task.
The IDE does not see the server
Open MCP settings from the IDE gear menu, confirm it is reading the expected config.toml, then restart or reload the IDE.
Wrong endpoint
Use the scoped Agiflow connection URL for coding agents. Do not use /api/v1/mcp-app/0.0.1; that endpoint serves the ChatGPT app/widget MCP surface.

Try Codex with Agiflow

Once Agiflow is connected, ask Codex to work from the board:

  • “What projects are in my Agiflow workspace?”
  • “Summarize the active work units for this project.”
  • “Turn this implementation plan into Agiflow tasks.”
  • “Start a workflow for this task and keep the board updated.”
  • “Add a comment with the decision we just made.”

You might also use

CursorBy Anysphere

Give Cursor your project board while it writes code.

Claude CodeBy Anthropic

Let Claude Code work your board tasks from the terminal.

AntigravityBy Google

Give Antigravity your board so it plans and writes code.

Browse all integrations

Build your own MCP connection

Agiflow exposes project work over MCP so any MCP-compatible client can use the same board context as your team.

View MCP connection docs