Connect your assistant
to DataEase AI MCP.

What do you need before connecting?

You need an eligible DataEase AI workspace, a personal token and an assistant that supports remote MCP over Streamable HTTP with custom bearer headers. Growth and Business include access by default. The server exposes 23 read tools and 1 action for recording completed fixes.

Check access in Account Settings → Connections → DataEase MCP. Workspace access settings can override the plan defaults. Starter and Pay As You Go do not include MCP by default; see current plans.

Your assistant must let you set an Authorization header. OAuth-only connection flows are not supported. If a client only offers an OAuth sign-in button, use a client or connection mode that supports custom headers.

This connection lets an external assistant use your DataEase AI data. It is separate from adding an external MCP server to DataEase AI's own server registry. The setup below is for connecting to DataEase AI.

How do you connect your assistant to DataEase AI?

Connect in 4 steps: create a token in the correct workspace, save its one-time secret, add the remote endpoint and bearer header to your assistant, then list your brands. Use the returned brand ID for subsequent reads so each request targets the intended brand.

1. Create a personal token

   In the standalone Brand Intelligence application, choose the workspace you want your assistant to access. Open Account Settings → Connections → DataEase MCP and select Create token. Give it a recognizable label, such as “My research assistant”. Choose an expiry of 30, 90 or 365 days; the default is 90.

2. Save the token while it is visible

   Copy the secret before closing the one-time display. Store it in your assistant's secure credential or environment settings. Do not put it in a prompt, shared document, screenshot or committed configuration file. If you lose it, create a replacement and revoke the old token.

3. Add a remote MCP connection

   Use the endpoint displayed in Connections. The production endpoint and required header are shown below. These are connection fields, not a universal configuration file; client interfaces differ.

   Transport

   Streamable HTTP

   Server URL

   https://mcp.dataease.ai/mcp

   Header name

   Authorization

   Header value

   Bearer YOUR_DATAEASE_MCP_TOKEN

   Replace YOUR_DATAEASE_MCP_TOKEN with your token, retaining Bearer and the space after it. Save or reconnect the client so it discovers the server's tools.

4. Check the connection with a brand list

   Ask your assistant: “Use dataease_list_brands to list the brands in my workspace. Show each brand's name, domain and ID, then let me choose one.” This reads saved data. If there are no brands, check the selected workspace and add your brand in the application.

Every subsequent brand request needs an explicit brandId from that list. A token cannot access another workspace's brands, and an incorrect ID does not silently select your primary brand.

What can you ask your connected assistant?

Start with 1 brand and a concrete decision: which buyer question needs attention, which existing page should improve, or what changed in referral traffic. Ask for dates, sources and sample counts. Saved scans, drafts and traffic reports can inform a draft you review.

Find the next website improvement

For the brand I selected, review the latest saved scan, prompt answers and open recommendations. Identify one missed buyer question, show the supporting answer and cited URLs, and suggest which existing page to improve. Do not run a new scan or mark anything fixed.

Review saved blog work

List this brand's saved blog posts and content ideas. Separate drafts from published records, read the most relevant saved draft, and suggest revisions. Report any missing content or continuation you still need to read.

Understand traffic without mixing metrics

Compare Search Console performance for the last 28 complete days with the previous period. Report GA4 sessions and recognized AI referrals separately. Show the date range, sample or row limits, cache fetch time and timezone where available.

A useful loop is discover → decide → create → verify: inspect saved evidence, choose a change, ask your assistant to prepare a draft, and review the result. Content generation or website edits performed by your assistant are separate from the DataEase AI MCP server. Use the platform's AI Agents for its agent workflows.

Which tools read brand intelligence?

The 7 intelligence tools expose your own brands, saved scans, prompts, answers, competitors and recommendations. Mention rates describe the selected sampled questions rather than total market visibility. Responses retain observation dates, and missing summaries or empty results are not replaced with invented data.

These reads do not start new scans or warm missing AI summaries. For background on interpreting results, read the founder's monitoring guide. Brand Readiness, Web Presence and AI Visibility remain the 3 components of Brand Presence Intelligence.

Which tools read saved content?

The 6 content tools retrieve saved ideas, blog posts and page artifacts, including drafts and published records. Generation status and publication status are separate. Listing a saved draft does not mean it is unpublished on every external website, and reading it never publishes it.

Markdown is the default detail format; HTML or stored block JSON can be requested explicitly. The server does not execute HTML or fetch embedded resources. Saved content may differ from an earlier published revision.

Blog Creator and generated page artifacts can be read independently of hosted Sites. Hosted records require Sites access and existing saved records. These tools do not make Pages generally available: Pages remains coming soon.

Which tools read traffic and search performance?

The 10 traffic tools cover source readiness, GA4, Search Console, crawler requests and hosted views. Google reads use the token creator's connection in the same workspace and the brand's selected property. GA4 sessions, search clicks, bot requests and hosted views are different measurements.

Connect Google and choose the brand's GA4 property or Search Console site in DataEase AI first. The assistant cannot supply an arbitrary property ID or borrow a connection from another workspace. Reads may contact Google on demand and reuse existing caches; a missing connection or provider error is not reported as zero traffic.

Date ranges use inclusive YYYY-MM-DD values, span at most 180 days and cannot extend into the future. Defaults are the last 30 complete days for GA4, 28 days ending 3 days ago for Search Console, and 30 days including today for stored traffic. GA4 scope defaults to all traffic.

GA4 has a 2-hour cache and Search Console a 5-minute cache. Reports identify source, period, cache state, fetch time and timezone when available. Fetch time is not Google's ingestion freshness. AI referrals cover recognized referrers, not every AI-influenced visit.

How do you record a recommendation you have already fixed?

Use the 1 write tool, dataease_mark_recommendation_fixed, after the work is complete and you have confirmed the details. It records the fix, moves the recommendation to Implemented and captures a baseline for tracking. It does not make the change, publish content or verify an improvement.

First obtain the recommendation ID from dataease_list_recommendations. Have your assistant show the selected brand, recommendation, fix kind, target URL, note and actual live date for your confirmation before calling the tool.

• new_url: a new page is live. Requires a target URL on the brand's own domain.
• updated_page: an existing page has been changed. Also requires an own-domain target URL.
• offsite: a listing, review, backlink or forum post. A reference URL is optional.
• other: other completed work. A reference URL is optional.

The optional note can contain up to 500 characters. markedAt is the date the fix went live, in YYYY-MM-DD format; it defaults to today and must be within the last 2 years, not in the future. Use the deployment date for website changes, not the date a draft was written.

Re-marking the same recommendation updates its existing fix record. Existing tokens can use this capability without being reissued. Undo and one-page re-audit verification stay in the application; verification can cost credits. Recording a fix itself uses no AI credits and does not prove that a citation, ranking or conversion improved.

How do you read a large list or a long article?

Use the continuation fields returned with each result. Intelligence lists expose nextOffset; content lists return a cursor. Long content includes a next offset and version. Follow those values until completion instead of treating the first page or chunk as the full result. Content lists default to 25 items.

Content lists allow at most 100 items per request. Prompt and recommendation catalogs retain a 200-entry ceiling. Ranked traffic reports default to 25 rows and cap at 100. A returned page can contain fewer rows than requested to fit the response limit, so use its continuation value rather than calculating your own.

For a long content body, pass the returned nextOffset as offset with the returned version. Offsets are UTF-16 positions, not byte counts. If the content version changes, restart the read. Use list-provided references for content details, not a file path or URL.

Who can create tokens, and how do you revoke access?

Every workspace member can create and revoke personal tokens. Owners and admins can also list and revoke all workspace tokens. Choose 30, 90 or 365 days of access. Revocation, expiry, membership and workspace eligibility are checked on requests, so an old token does not preserve removed access.

A token is personal to its creator and bound to one workspace. It can read accessible saved brand data and content, including drafts, and record completed fixes. Use a separate token for each assistant connection so you can recognize and revoke it independently.

To disconnect, return to Account Settings → Connections → DataEase MCP, find the token and choose Revoke, then confirm. Token listing and revocation remain available after a downgrade. To rotate a token, create a replacement, update the client and revoke the old one.

What should you check if a connection or tool fails?

Check 3 things first: the exact endpoint and bearer header, the token's status and workspace access, and any Google connection required by the requested tool. Errors remain explicit. A failed request is not an empty report, and retrying cannot restore a revoked token or missing permission.

The client insists on OAuth sign-in

Use a remote Streamable HTTP connection with custom headers. OAuth-only clients are not supported by this server.

Unauthorized, expired or revoked token

Check that the header contains Bearer, a space and the current secret. Check token status, membership and access in Connections. Create a replacement if the secret is lost or expired.

MCP access is unavailable

Check the selected workspace and effective access settings. Growth and Business include access by default. A workspace override may change eligibility.

A brand, recommendation or saved item is not found

List brands again with this token, choose the correct brand ID and obtain the recommendation ID or content reference from that brand's list. IDs from other workspaces are not accepted.

A Google report fails

Reconnect Google as the token creator in this workspace, check the selected brand mapping and granted scopes, then retry. Google permission, quota and timeout errors need to be resolved at their source.

A saved list is empty

Check brand, source and status filters. Some legacy artifacts cannot be assigned to a brand. Hosted content also requires Sites access. An empty list does not trigger content generation.

A rate limit or timeout occurs

The server allows 120 requests per token per minute. Reduce parallel requests and retry with backoff. Narrow a report or read content in chunks when a result is large.

The record-fix tool is listed but returns an unknown-tool error

Tool discovery alone does not verify the action's backend availability. Contact DataEase AI support and use Opportunities in the application to mark the fix meanwhile. Do not keep retrying the write.

What else should you know about DataEase AI MCP?

These 6 answers cover the boundaries most likely to affect your setup: scans and credits, write access, compatible clients, drafts, Google connections and hosted content. Check them before choosing a connection so your assistant has the access and capabilities you actually intend.