Learn

Search Console API, BigQuery and MCP: three ways to get your data out

The Search Console interface stops at 1,000 rows. These are the three documented ways to pull more: the API, the BigQuery export, and an MCP server that lets an AI client use either, with the limits and a working request.

By Adem ÇakalPublished 6 sources7 min read
Short answer

There are three routes. The Search Analytics API returns up to 25,000 rows per request and pages with startRow, but Google says it returns top rows, not guaranteed all data. Bulk data export to BigQuery writes daily tables for ongoing history but does not backfill. An MCP server is a wrapper that lets an AI client call one of the other two; choose a read-only scope and a server you can audit.

The three routes compared

RouteBest forHistoryCost and setup
Search Analytics APIScripts, reports, one-off pullsThe route Google points to for data before an export was set upFree; needs a Google Cloud project and OAuth
Bulk export to BigQueryLong-term storage, joins, large sitesStarts at setup; first export within 48 hoursBigQuery storage and query costs, with a free usage level; billing required
MCP server over the APIAsking an AI client questions about your dataSame as the API underneathDepends on the server; you also trust its code
Data path: Search Console feeds the Search Analytics API and the BigQuery export; an MCP server sits in front of either and an AI client talks to the MCP server Search Console Search Analytics API BigQuery export MCP server AI client
The MCP server is optional: scripts can read the API or BigQuery directly.

Route 1: the Search Analytics API

The request is a POST to the searchAnalytics/query method for one property. Dates are in Pacific Time. Grouping dimensions are applied in the order you list them, and rows come back sorted by clicks. Here is a request body that pulls pages and queries for one month, only for pages under /blog/:

{
  "startDate": "2026-10-01",
  "endDate": "2026-10-31",
  "dimensions": ["page", "query"],
  "type": "web",
  "dimensionFilterGroups": [{
    "groupType": "and",
    "filters": [{"dimension": "page", "operator": "contains", "expression": "/blog/"}]
  }],
  "rowLimit": 25000,
  "startRow": 0
}

Send it with an OAuth access token. The property in the URL must be encoded, for a Domain property written as sc-domain%3Aexample.com:

curl -X POST \
  "https://www.googleapis.com/webmasters/v3/sites/sc-domain%3Aexample.com/searchAnalytics/query" \
  -H "Authorization: Bearer $GSC_TOKEN" -H "Content-Type: application/json" \
  -d @body.json

To page past the first 25,000 rows, repeat the request and raise startRow until a response returns fewer rows than the limit:

import os, requests
SITE = requests.utils.quote("sc-domain:example.com", safe="")
URL = f"https://www.googleapis.com/webmasters/v3/sites/{SITE}/searchAnalytics/query"
HEAD = {"Authorization": "Bearer " + os.environ["GSC_TOKEN"]}
rows, start = [], 0
while True:
    body = {"startDate": "2026-10-01", "endDate": "2026-10-31",
            "dimensions": ["page", "query"], "rowLimit": 25000, "startRow": start}
    r = requests.post(URL, json=body, headers=HEAD)
    r.raise_for_status()
    batch = r.json().get("rows", [])
    rows += batch
    if len(batch) < 25000:
        break
    start += 25000

This is an example, not a full client: obtaining the token and refreshing it are separate steps in Google's authorization documentation.

Route 2: bulk export to BigQuery

Search Console can export to a BigQuery dataset in your own Google Cloud project. Google's help describes three tables: one aggregated by property, one aggregated by URL, and a log of successful exports. Exports run once per day, though not necessarily at the same time for each table. Rare queries are flagged as anonymized, and their query field is empty. Google's reference page describes that field both as null and as a zero-length string, so test which one your dataset contains before writing filters. We did not find a stated row limit or a comparison with the API in the pages we read.

Two practical points from the setup page: the dataset location is hard to change once exports begin, and schema changes cause the export to fail, so do not alter the exported tables.

Limits at a glance

ItemLimitSource page
Rows per API request1 to 25,000; default 1,000Search Analytics reference
PagingstartRow, zero-basedSearch Analytics reference
CompletenessTop rows returned; not guaranteed to be all dataSearch Analytics reference
Queries per minute, same site or user1,200 eachUsage limits page
Queries per day and minute, same project30,000,000 and 40,000Usage limits page
Load quotaMeasured per 10 minutes and per day; no numbers publishedUsage limits page
URL Inspection, per site2,000 per day, 600 per minuteUsage limits page
Bulk export frequencyOnce per day per tableBulk export reference
Bulk export historyStarts at setup; use API for earlier dataBulk export setup page

Route 3: MCP servers that wrap the API

The Model Context Protocol is described on modelcontextprotocol.io as an open-source standard for connecting AI applications to external systems, such as data sources and tools. An MCP server for Search Console exposes API calls as tools that an AI client can use, so you can ask a question in plain language and the server runs the query. We do not endorse a specific third-party server; many exist and we have not audited them.

The security page on the same site covers local server compromise, token passthrough and scope minimization. Applied here:

  • Request the read-only scope. The API accepts https://www.googleapis.com/auth/webmasters.readonly as well as the full webmasters scope. Reporting needs only the first. Do not approve write access for a server that only reads data.
  • Treat the server as code you run. A local server executes with your privileges. Read its source, pin its version, and prefer one that does not need shell access.
  • Store tokens carefully. A leaked token with a broad scope reaches every property you can access. Revoke tokens in your Google account when you stop using a server.
  • Mind the quotas. An agent that loops can burn per-minute limits quickly; ask it to request only the dimensions and dates it needs.

With the pulled data, our average position explainer helps you read the position column, and the Search Console AI report guide covers a view that is not in the API.

Questions people ask

Why does my API pull return fewer rows than the interface shows?

Google says the API returns top rows and does not guarantee all data. Narrow the date range or add filters and run several smaller requests.

Can I backfill BigQuery with old data?

Google's setup page does not describe a backfill. It points to the API or reports for data from before the export began, so start the export early.

Which OAuth scope should an MCP server use?

For reporting, the read-only webmasters.readonly scope. The API also accepts the broader webmasters scope, which you should withhold unless you need to change settings.

Does the API include the new Generative AI report?

The reference page we read lists web, discover, news, image and video search types and does not mention it. Check the current reference before assuming it is available.

Sources

  1. Google Search Console API: searchAnalytics.query reference, accessed 2026-11-12.
  2. Google Search Console API: Usage limits, accessed 2026-11-12.
  3. Search Console Help: Set up bulk data export, accessed 2026-11-12.
  4. Search Console Help: Bulk data export table guidelines and reference, accessed 2026-11-12.
  5. Model Context Protocol: What is MCP?, accessed 2026-11-12.
  6. Model Context Protocol: Security best practices, accessed 2026-11-12.