API Keys and AI CLIs
Connect hosted providers, local AI CLIs, and MCP integrations without sending your workspace data to a 1MarketingTool cloud.
1MarketingTool is BYOK. You decide which providers to connect, and the app uses your own API keys or local AI CLIs.
Supported AI CLIs
The app drives local AI command-line tools you already have installed. Each is detected on your machine; enable only the ones you actually use.
| CLI | Image generation |
|---|---|
| Claude Code | No |
| Codex | Yes |
| Gemini | Yes |
Antigravity (agy) | Yes |
| OpenCode | No |
| Amp | No |
| Qwen | Yes |
| Hermes Agent | No |
| Aider | No |
| Grok | No |
Pick the CLI and its model in the AI CLI picker in the top bar. You can also install the SEO agent skills into any of them.
Only Codex, Gemini, Qwen and Antigravity ship a first-party image model, so only those can run Image Generation on the Local AI route. Choosing one of the others there is refused up front with the reason.
A local CLI has no web retrieval. Where that matters — such as AI Visibility trackers — its answers are labelled as model memory rather than a live citation.
Kiro CLI and Claude Desktop are MCP clients, not AI CLIs the app drives. You install 1MarketingTool's MCP server into them; see MCP Integrations.
Hosted Providers
Use hosted providers when a workflow needs external data or a model you prefer:
- OpenAI, Claude, Gemini, or another AI provider for generation.
- DataForSEO for keyword, SERP, backlink, and ranking data.
- Google Search Console and analytics services for site performance.
- Social channel APIs for scheduled publishing.
Models on Your Own Machine
You can point the BYOK route at a model running on this computer or on your local network instead of a hosted service. This works with anything that answers in the OpenAI shape, such as LM Studio, Ollama, or 1LocalAITool, and with anything that answers in the Anthropic shape, which 1LocalAITool also does.
- Open Settings → AI and switch to BYOK.
- Pick Custom OpenAI-compatible under Gateways, or Anthropic under Protocols if your server speaks the Anthropic shape.
- Set Base URL to the address your local server prints, for example
http://localhost:1234/v1for LM Studio orhttp://localhost:11434/v1for Ollama, and set Model to the model id it serves. - Paste something into API key. The field cannot be empty, but a local server that does not check keys accepts any value.
- Click Test connection, then Use as active.
A model on a CPU or a small GPU writes slowly, so 1MarketingTool gives a local Base URL up to ten minutes to answer, where a hosted provider gets 45 seconds. That longer limit applies when the Base URL host is localhost, a 127.x, 10.x, 192.168.x or 172.16–31.x address, a .local name, or a Tailscale address (100.64–127.x or a .ts.net name). It applies to Custom OpenAI-compatible since 1.5.0 and to Anthropic since 1.5.1. A long campaign or article from a slow local model finishes instead of timing out halfway.
Key Storage
API keys and tokens should be stored through the app settings. Publishing tokens are stored in the operating system keychain where supported. Avoid pasting secrets into campaign prompts or content fields.
MCP Integrations
1MarketingTool exposes its rank, keyword, content, video, and browser tools through a single local onemarketingtool MCP server. Use it when you want compatible AI clients to reach app-specific workflows from your machine — see MCP Integrations.
Troubleshooting
If a provider fails, check the exact error in Logs or the relevant settings screen. Most failures are caused by missing API permissions, expired tokens, or a CLI that is not on the system path.