Let your AI agent publish — but only what you approved

You already do half your marketing inside a coding agent. You ask Claude Code to draft the launch thread, you ask Codex to summarise the release notes, and it does a decent job. Then the handoff starts: copy the text out of the terminal, open the scheduler, paste it into the right channel, attach the image, pick a time. The agent wrote the post in ten seconds and you spent five minutes moving it around.

The obvious fix is to let the agent post for you, and the obvious fear stops you. An agent with posting rights is one bad prompt away from publishing a half-finished draft to your brand account. So you keep copying and pasting, because a slow manual step feels safer than a fast one you cannot supervise.

What changed

1MarketingTool 1.3.0 gives your AI tools a single, supervised way in. A new AI Access page connects Claude Code, Codex, Cursor and the other coding agents to the app's local onemarketingtool MCP server in one click, and a new Publishing tool group lets those agents post and schedule your library content. You decide how far that goes. Set the guardrail to Approved content only and an agent can publish nothing you have not approved yourself.

How it works in practice

Connect the agent you already use

AI Access used to be a sub-tab buried in Settings. It is now its own page, and the top of it tells you whether the setup is working before you go looking for problems: the server state and port, Connected tools, Tools exposed, Calls · 24h, Failed · 24h and Last call. One master switch, Allow AI tool calls, turns every agent off at once.

The Setup tab lists every supported client with a detected state: Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor, Windsurf, VS Code, Claude Desktop, Cline, Kiro, Hermes Agent, Antigravity CLI, Kimi Code, Roo Code, Qoder CLI, Trae, Factory Droid, Warp, Augment, Pi, Devin CLI and more. Connect writes a onemarketingtool entry into that tool's own MCP config. Restart the tool and the tools appear as mcp__onemarketingtool__*. If you would rather edit the config yourself, the Manual menu shows a copy-ready snippet that already contains this machine's real launch command, so you do not have to hunt for paths.

From then on you can talk to the app by name inside your agent: "use 1mkt to list my projects" works, and so do onemarketingtool and 1marketingtool.

Decide what each agent may touch

The Tools & permissions tab splits the server's 34 tools into groups you can switch on and off: Rank & keywords, Content, Video, Website capture and Publishing. Turning a group or a single tool off hides it from agents you start afterwards and refuses calls from agents that are already running. Below the switches, a searchable tool reference labels every tool Read, Writes or Publishes, so you know which ones can post in public before you enable them.

A refused call is not a dead end for the agent. The refusal names the exact switch to flip. It tells the agent to ask you to enable "Publishing" in 1MarketingTool → AI Access → Tools, so the agent can pass that on to you instead of retrying blindly.

Approve what goes out, then let the agent post it

The Publishing group gives agents five tools: list the channels that can post right now, publish a library piece, schedule one for later, list scheduled and published posts, and cancel a scheduled post. Under the switch you choose one of two guardrails:

• Approved content only. Agents can post or schedule only the pieces a person approved, and they post the approved text. If an agent tries to swap in its own wording, the call is refused, and the agent is told to save the new text as a draft for you to approve.
• Any content. Agents can also publish drafts they just wrote and replace the post text. This is the default, so if agents will be posting to accounts that matter, switch to Approved content only before you connect them.

Approval stays a human step, and there are two ways to give it. Open a piece and click Approve. Any unsaved edits are saved first, so the approval covers exactly what is on screen. Or select several drafts in the Content Library and use the bulk Approve action, which only ever moves drafts forward and never pulls a scheduled or published piece back.

Now the prompt is just the prompt. Ask your agent to publish the latest campaign and it lists your channels, picks the approved pieces and posts them.

Agent posts do not take a shortcut. They run through the same post-now and schedule tasks as the Campaign Results page, so they show up in the Activity Center, failures reach your alert inbox, and project channel mapping and paused channels still apply. A scheduled post goes out when it is due, as long as the app is running.

See every call afterwards

The Activity tab records every call an agent makes under the name of the tool that ran, and you can filter it down to Failed or Blocked. Calls your permission switches refused show as blocked rather than failed, so a guardrail doing its job does not look like an error. The same calls also land in Logs → AI.

The local bridge that agents talk to is locked down as well. Every call has to carry a token that is created each time the app launches and can only be read by local processes, and requests that come from a web page are refused. A website you have open cannot quietly drive your publishing tools.

Before vs After

Step	Before	After
Connect an agent	Find the MCP settings sub-tab, copy a config block, fix the paths by hand	Click Connect on AI Access, restart the tool
Get a draft out	Copy text from the terminal, paste it into the scheduler	The agent calls the publish or schedule tool
Keep bad drafts off your account	Do not let agents near posting at all	Approved content only: agents post only what you approved
Approve a batch	No bulk action	Select drafts in the Library and click Approve
Limit what an agent can do	Hand-edit an environment variable in each client config	Per-group and per-tool switches, applied even to agents already running
Audit what happened	Scroll the agent's own transcript	The Activity tab, filterable to failed and blocked calls

Who benefits most

Solo founders shipping every week. You already ask an agent for the launch copy. Now you approve it once in the Library and let the same agent schedule it across channels, without the copy-paste step between them.

Anyone running agents with bypass permissions. Coding agents in auto-approve mode will call whatever tools they can see. The group switches and the Approved content only guardrail give you a limit that sits in the app, not in the agent's prompt.

Teams posting from a brand account. Also new in 1.3.0: the X channel has a Connect button. Approve the app in a browser signed in to the account you want to post as, such as your brand handle instead of the developer account that owns the API keys, and the tokens are stored on this device. The setup fields now match the current X console (Consumer Key and Consumer Key Secret), and an out-of-credits error now points you to the X billing page that actually holds your credits.

Try it

Update to 1MarketingTool 1.3.0 and open AI Access. Connect the agent you already use and set Publishing to Approved content only. Then approve a few drafts in the Content Library and ask the agent to post them. The agent does the busywork, and nothing reaches your audience until you have approved it.