MCP client

AgentSEO + Claude Desktop

Want Claude Desktop to act more like an SEO research assistant than a chat box? Add the local AgentSEO MCP package, ask one focused question, and get structured evidence plus a plain-language explanation back in the same session.

From a social thread? Copy the setup first, validate one live workflow, then decide whether it belongs in docs, automation, or your agent runtime.

Local MCP packageDesktop workflowPrompt-driven research

Copy Claude Desktop setup Get free API key

Why this page converts

Best for analysts, founders, and operators who want fast one-off research before they automate anything heavier.

Core workflows

What this page should convert

Run content-gap or local-audit checks from a desktop chat instead of switching between dashboard tabs.
Inspect a SERP and ask Claude to explain the evidence in plain language after the tool call.
Use AgentSEO as a research assistant for one-off analyses before codifying them into automations.

Setup snapshot

Claude Desktop MCP config

{
  "mcpServers": {
    "agentseo": {
      "command": "npx",
      "args": ["-y", "@agentseo/mcp-server"],
      "env": {
        "AGENTSEO_API_KEY": "sk_live_...",
        "AGENTSEO_API_URL": "https://www.agentseo.dev/api/v1",
        "AGENTSEO_PROJECT_ID": "client-alpha",
        "AGENTSEO_WORKFLOW_ID": "desktop-research"
      }
    }
  }
}

Lead with the local MCP package. Anthropic currently treats Claude Desktop remote connectors differently from Claude Code CLI MCP setup, so this page should avoid implying a copy-paste hosted MCP path unless you test it directly in Claude Desktop.

Implementation plan

How to set it up

1
Install the local @agentseo/mcp-server package through an npx-based MCP config.
2
Set AGENTSEO_API_KEY and optional attribution defaults in the Claude Desktop MCP server environment.
3
Restart Claude Desktop so the MCP tools are available in the session.
4
Ask Claude to run a narrow AgentSEO task first, such as a content gap or local audit, before expanding scope.

Output shape

What users get back

Direct MCP tool access without building a wrapper app.
Readable research flow for teams that want conversational analysis on top of structured output.
The same deterministic AgentSEO result shapes used by other runtimes.

Limits

What to say plainly

Best documented path today is local MCP, not a guaranteed hosted remote connector flow.
Desktop chat is less operational than a workflow runner or agent runtime.
Large repeated jobs still belong in scheduled systems, not ad hoc chat sessions.

Visual flow

Claude Desktop research loop

Claude Desktop is a good front end for one-off research before the workflow graduates into automation.

Desktop Chat

Analyst asks a narrow SEO question.

Local MCP

Claude calls the local AgentSEO MCP server.

Structured Result

Tool returns JSON-shaped evidence.

Human Review

Claude explains the result in plain language.

Concrete example

Ad hoc content-gap research in Claude Desktop

Use Claude Desktop as a research surface when you want AgentSEO data plus an immediate plain-language explanation for a page opportunity.

Example prompt

Run AgentSEO content gap analysis for https://example.com/seo-audit
against the keyword "seo audit checklist".
Summarize the missing topics, likely intent, and the next three updates worth making.

Expected tool flow

Claude calls:
1. agentseo_content_gap

Then summarizes:
- missing topics
- page intent
- act_now vs monitor
- recommended updates

This is strongest for exploratory work where the user wants both structured output and a natural-language synthesis.
Once a workflow becomes repeatable, move it into n8n, Make, Claude Code, or another system with better orchestration.
Use narrow prompts first so the tool call stays scoped and reviewable.

FAQ

Questions this page should answer fast

Package-level docs

Should Claude Desktop and Claude Code use the same page?

No. The user intent and setup flow differ enough that they deserve separate pages, even though both use MCP.

Why focus on local MCP for Claude Desktop?

Because Anthropic's current help docs separate Claude Desktop connector setup from Claude Code CLI MCP setup, so local MCP is the clearer documented path for AgentSEO right now.