Hosted MCP + local npm

Give your AI agent live SEO tools—not another tab to read

Connect AgentSEO to Claude Code, Codex, Cursor, or another MCP client. Your agent can inspect live search evidence, choose the next SEO workflow, and return structured output in the session where you are already working.

Get free API key View connection options

45 tools discovered livePublic catalogKey-protected execution

Hosted connectionagentseo.mcp

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

First useful prompt

Find content gaps for example.com around "seo api for ai agents". Use live search evidence, then turn the best opportunity into a brief with an outline and next actions.

Choose the connection

Hosted for the shortest path. Local when the client needs stdio.

Both paths call the same AgentSEO workflow layer. The difference is where the MCP transport runs—not what the SEO data can do.

Hosted Streamable HTTP

No local wrapper. Use the production endpoint and pass your key as a private header.

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

Keep API keys out of repositories and shared project configuration. Optional project and workflow headers improve attribution.

Local npm stdio

Use the published package when the host launches a local MCP process.

{
  "mcpServers": {
    "agentseo": {
      "command": "npx",
      "args": ["-y", "@agentseo/mcp-server"],
      "env": {
        "AGENTSEO_API_KEY": "sk_live_your_key",
        "AGENTSEO_API_URL": "https://www.agentseo.dev/api/v1"
      }
    }
  }
}

What the agent can do

Move from a question to evidence and a next action

MCP is useful when the model needs to inspect live state before it recommends something. For scheduled or backend execution, keep using the REST API.

Compare SDK and MCP paths

SERP and keyword research

Inspect live search results, SERP features, keyword ideas, ranking pages, and competing domains before the model recommends a direction.

Content planning and QA

Build briefs, outlines, schema plans, title options, internal-link suggestions, and pre-publish checks from one tool surface.

Backlink and competitor analysis

Review referring domains, anchor patterns, competitor intersections, lost links, and bounded outreach opportunities.

Local visibility and rank tracking

Run local audits, track visibility across locations, monitor rankings, and poll longer-running jobs inside the same agent loop.

AI search visibility

Build repeatable prompt sets and inspect AI Overview signals without asking the model to guess from page copy alone.

Scoped, attributable calls

Keep execution behind an API key and attach optional project and workflow IDs so usage can be traced to the calling job.

Verify before you connect

The public records match the live product surface

These records verify distribution and catalog discovery. They do not, by themselves, prove output quality or reliability for your workflow—test one bounded use case before putting it into production.

45 discoverable tools

Hosted endpoint

The live server card publishes Streamable HTTP transport and the current tool catalog. Tool execution requires an AgentSEO API key.

Inspect server card

Latest: v0.1.4

Official MCP Registry

The official listing records the hosted endpoint and the npm distribution under dev.agentseo.www/agentseo-mcp.

Verify registry record

Package: v0.1.3

npm package

Use @agentseo/mcp-server when an MCP host needs a local stdio process instead of the hosted endpoint.

View npm package

Before the first call

Practical MCP questions

Should I use the hosted MCP endpoint or the local npm package?

Use the hosted Streamable HTTP endpoint when you want the shortest setup and no local wrapper. Use the npm package when your MCP host or operating environment needs a local stdio process.

Does AgentSEO work with Claude Code, Codex, and Cursor?

AgentSEO exposes standard MCP transports for compatible hosts, including Claude Code, Codex, Cursor, Claude Desktop, and custom MCP clients. The exact setup screen or command varies by client.

Do I need an AgentSEO API key?

Yes for tool execution. Discovery methods such as initialize and tools/list remain public so compatible clients and directories can inspect the server. Calls that run SEO work require an API key and consume credits according to the underlying endpoint.

What should I run first?

Choose one bounded job with an observable output. A content-gap check, SERP outline, or local audit is easier to review than an open-ended request to improve an entire site.

When should I use REST instead of MCP?

Use REST for backend services, scheduled jobs, queues, and integrations where your code controls the sequence. Use MCP when a compatible agent should discover and call tools during an interactive workflow.

Start with one workflow

Connect the server. Test a real SEO decision.

Create a key, choose hosted or local MCP, and run one reviewable task before you automate a larger workflow.

Get free API key Read quickstart