MCP client

AgentSEO + Claude Code

Want Claude Code to stop scraping pages and start using real SEO tools? Add AgentSEO over MCP and let Claude inspect a SERP, branch into the next check, and return a usable answer inside the editor.

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

Official MCP pathTool-callingDeveloper workflow
Copy Claude Code setupGet free API key

Why this page converts

Best for builders who want copy-paste MCP setup now and a stronger agent loop than raw scraping or manual browser research.

Core workflows

What this page should convert

  • Let Claude inspect a SERP and branch into content gap or local audit workflows.
  • Run repeatable prompt-driven research with deterministic tool outputs.
  • Keep project and workflow attribution attached to every tool call.

Setup snapshot

Claude Code MCP command

claude mcp add --transport http agentseo https://www.agentseo.dev/mcp \
  --header "Authorization: Bearer sk_live_..." \
  --header "x-project-id: client-alpha" \
  --header "x-workflow-id: nightly-refresh"

This page can confidently lead with MCP because the product already ships both hosted and local setup options.

Implementation plan

How to set it up

  1. 1

    Generate a server-side AgentSEO API key with no allowed browser domains.

  2. 2

    Add the hosted MCP endpoint to Claude Code, or use the local @agentseo/mcp-server package.

  3. 3

    Restart Claude Code so the MCP server is available in the tool list.

  4. 4

    Prompt Claude to run a specific AgentSEO tool such as local audit, content gap, or AI overview extraction.

Output shape

What users get back

  • Direct tool access for SERP analysis, local audits, content gap checks, and job status lookups.
  • Agent-native response shapes designed for deterministic follow-up prompts.
  • A cleaner alternative to manual scraping or oversized provider payloads.

Limits

What to say plainly

  • Hosted MCP requires a server-side key without allowed domains.
  • Users still need to choose the right workflow and prompt structure.
  • MCP is powerful, but not every automation belongs inside an editor.

Visual flow

Claude Code tool loop

Claude Code is strongest when the model reads one tool result, decides, and calls the next tool without leaving the editor.

1

Prompt

User asks for SERP triage or page opportunity analysis.

2

MCP Tool

Claude calls analyze_serp or another AgentSEO tool.

3

Branch

Model decides content gap, local audit, or stop.

4

Summary

Claude returns evidence plus next action.

Concrete example

SERP triage inside Claude Code

Ask Claude to inspect a keyword, decide whether the opportunity is editorial or local, and call the next AgentSEO tool without leaving the editor.

Prompt flow

Analyze "best seo api for ai agents" with AgentSEO.
If the SERP looks editorial, run content gap against https://www.agentseo.dev/.
If the query looks local, switch to local audit instead.
Return the decision, evidence, and next action.

Expected outcome shape

Claude calls:
1. agentseo_analyze_serp
2. agentseo_content_gap (if editorial)

Then summarizes:
- likely intent
- competition level
- missing topics
- recommended next call
  • This is where MCP shines: the model can inspect a tool result and branch instead of forcing you to copy JSON by hand.
  • Keep prompts concrete about the branch condition so the tool chain stays deterministic.
  • Use projectId and workflowId headers when you want attribution grouped by client or workflow.

FAQ

Questions this page should answer fast

Package-level docs

Should Claude Code and Claude Desktop share one page?

No. Claude Code has different intent and setup behavior, even though both rely on MCP.

When should users pick local MCP instead of hosted MCP?

Use local MCP when a key is restricted to browser origins or when the team prefers the MCP process on its own machine.