AI Access (MCP)

AI Access connects the AI tools you already use (Claude Code, Codex CLI, Cursor and 20+ others) to 1MarketingTool's local MCP server. Once connected, an agent can check rankings, write a campaign piece, drive a browser capture, edit a video storyboard, or post your approved content, all without leaving its own terminal. Open it from the sidebar under System → AI Access.

The app must be running. Every tool call goes to the app over a local bridge, and nothing is exposed to the network.

The Status Card

The card at the top shows 1MarketingTool MCP server with its state (running and the port it listens on, Access off, or Bridge offline) and five numbers:

Allow AI tool calls is the master switch. Turn it off and every call from every agent is refused until you turn it back on. Refresh re-checks which clients are installed and connected.

Setup: Connect Your AI Tools

The Setup tab lists each supported client with a state: Connected, Needs update, Not connected, Not installed, Manual setup, or Config unreadable. Clients found on this machine are listed first.

• Connect writes a onemarketingtool entry into that client's own MCP config file (for example ~/.claude.json for Claude Code or ~/.codex/config.toml for Codex CLI). Restart the client afterwards so it loads the server.
• Disconnect removes the entry again.
• Manual (or Setup, for clients with no one-click option) opens copy-ready install snippets. They already contain this machine's real launch command and paths, so you can paste them as they are.

Supported clients include Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor, Windsurf, VS Code, Claude Desktop, Cline, Kiro, Grok Build, Hermes Agent, Antigravity CLI, Kimi Code, Roo Code, Qoder CLI, Trae, Factory Droid, Kilo Code, Warp, Augment, Pi, Oh My Pi, and Devin CLI.

In the client, the tools show up as mcp__onemarketingtool__*. You can also ask for them by name: "use 1mkt to list my projects". The names 1mkt, 1marketingtool and onemarketingtool all work.

If you update or move the app, saved entries that point at a server script that no longer exists are repointed to the current install the next time 1MarketingTool starts.

Why the server name has no digit. Some AI clients silently expose no tools when a server name starts with a number, so the server is called onemarketingtool rather than 1marketingtool.

Tools & Permissions

The Tools & permissions tab decides what agents are allowed to do. Turning a group or a single tool off hides it from AI tools you start afterwards and refuses calls from ones that are already running.

get_products is always available so agents can look up project IDs. To narrow a single client further than the switches allow, set ONEMARKETINGTOOL_MCP_GROUPS=rank,content (any comma-separated list of groups) in that client's MCP entry.

The Tool reference below the switches lists every tool with its group and access level: Read (reads data and changes nothing), Writes (creates or changes data inside 1MarketingTool), or Publishes (posts publicly to your connected channels). Search it or filter by group, and switch individual tools off.

When a call is refused, the agent gets a message naming the exact switch, such as enabling "Publishing" in 1MarketingTool → AI Access → Tools, so it can ask you rather than retry.

Publishing and the Approval Guardrail

The Publishing group lets agents list the channels that can post for a project, publish a library piece, schedule one for a future time, list scheduled and published posts, and cancel a scheduled post. When the group is on, pick one of two guardrails:

• Approved content only. Agents can post or schedule only pieces you approved, using the approved text. A request to publish a draft, or to replace the post text, is refused.
• Any content. Agents can also publish drafts they just wrote and replace the post text. This is the default. Use it only with agents you trust.

Approving a piece is always something a person does in the app. Open the piece and click Approve (pending edits are saved first, so the approval covers the text on screen), or select drafts in the Content Library and use the bulk Approve action. See Content Library.

Agent posts go through the same post-now and schedule tasks as the Campaign Results page. They appear in the Activity Center, failures reach the alert inbox, and project channel mapping and paused channels still apply. Scheduled posts go out when they are due, as long as the app is running.

Activity

The Activity tab lists every MCP call with the name of the tool that ran and its result: OK, Failed, or Blocked (refused by your permission switches). Filter the list to failures or blocked calls, or clear it. The same calls are also recorded in Logs → AI.

Skills

The Skills tab installs reusable writing, SEO, and design skills into the local CLI agents 1MarketingTool detected.

Safety

• The bridge that agents talk to only accepts calls carrying a token that is created each time the app launches and stored in a file only your user account can read.
• Requests that come from a web page are refused, so a site open in your browser cannot call the publishing tools.
• API keys, OAuth tokens and channel credentials stay inside the app. Agents call tools and never receive the credentials behind them.
• Turn on only the groups an agent needs, and prefer Approved content only for any agent that can publish.

Next

• API Keys and AI CLIs — the agents that connect here
• Publishing — how posts go out once they are queued
• API Providers — the keys these tools spend