Work with an AI agent

Everything in the app is also available to AI agents through OpenSEO's MCP server: research, saving, trackers, audits, AI visibility and local SEO. You ask in plain language; the agent picks the tools, checks prices and keeps under the ceiling you set.

Connect once

Create an API key under Settings and add the server to your client. The exact steps for Claude Code and other clients are in Connect an agent over MCP.

How agents handle money

Before the examples, the rules every OpenSEO tool follows:

• Tools that spend take dry_run, which returns the price and spends nothing, and max_cost_usd, a ceiling you agree with the agent. A call whose price could be higher is refused before anything is spent.
• Background runs (run_rank_check, run_ai_visibility, run_geo_grid) require a ceiling.
• Cached results are free, so asking the agent to look again at something from this week costs nothing.
• Your monthly spending limit applies to agents too. It is the backstop if you give an agent a broad task; set one under Settings before you do.

Three workflows

Research and save keywords

Using OpenSEO, find keyword ideas for "emergency plumber" in my project.
Keep it under $0.20. Save the 20 with the highest volume and difficulty
under 40, tagged "q4".

1. list_projects

   The agent finds your project. With only one project, it can skip this.

2. keyword_ideas with dry_run

   It checks the price of 150 ideas (about $0.08) against your $0.20.

3. keyword_ideas with max_cost_usd: 0.20

   It runs the lookup with your ceiling and filters the results.

4. save_keywords

   It saves the 20 keywords with their figures and the tag. Saving is free.

Set up tracking, with prices first

Create a weekly rank tracker for acme.com in Austin, Texas, on mobile,
top 20, with the keywords tagged "q4". Tell me what one check and one
month will cost before running anything.

The agent uses search_locations to find Austin's location code (free), list_saved_keywords with the tag, and create_rank_tracker, which spends nothing. list_rank_trackers then gives the price range of one check, and the agent multiplies it by about 4.35 for a weekly month. It runs run_rank_check only when you agree, with that range as its ceiling, and reads the outcome with get_rank_results.

Give the agent memory of your project

Read the project context. Then add our goal (more booked jobs in north
Austin), our two main competitors (globex.com, initech.com) and the
pages we care about most, and save it back.

Each project has free-form context notes that any agent can read with get_project and replace with update_project_context. Keep your goals, audience, competitors and key pages there, and every agent session starts from the same understanding.

More things to ask for

The full list of tools and what each one costs is in MCP tools.