Use Stride with your AI assistant

Connect your AI assistant to Stride to understand your visibility, manage Actions, create content and report work without losing its context.

Stride MCP lets your AI assistant use your approved Stride data and capabilities, then keep the resulting work connected to the site it belongs to.

Availability: Stride MCP is included with Scale and Agency. Agency access covers all sites in the account, including Starter and Grow client sites. Standalone Starter and Grow accounts do not include MCP. Your assistant can use only the sites and permissions you explicitly approve, within your current account and site role and existing plan allowances. Operational switches can temporarily disable the service or individual capabilities. This does not change the separate CMS or Shopify connector pilots.

Overview: what is Stride MCP?

MCP stands for Model Context Protocol. It is a way for an AI assistant to connect to another service and use its tools with your permission.

With Stride MCP, you can stay in your AI app while asking questions such as:

“Which buyer questions are we least visible for, and what evidence supports that?”
“What is the next useful Action for this site? Check what is already in progress first.”
“Use our brand rules, personas and approved sources to set up an article.”
“Let Stride prepare the research, but let my own model write the draft.”
“Save this draft in Stride so I can review and edit it there.”
“Record what we changed, and show what happened to visibility after it shipped.”

This is the Stride connection, not a separate CMS publishing connection. It does not give your assistant unrestricted control of your account, website or computer.

What can your assistant do?

Area	What your assistant can do
Understand visibility	Read saved visibility results, tracked questions, trends, competitors, citations, topics, opportunities, AI traffic and scan results.
Use your brand knowledge	Read brand information, approved facts and available Knowledge Base sources. With permission, update supported brand information and knowledge records.
Inspect and manage Actions	Read the queue, priority rationale, saved progress, blockers, projects and linked drafts. Resume existing work, adjust supported setup, save edits, hide or restore queue items, dismiss eligible saved work and change your personal board order.
Create and optimize content	Configure Content work and explicitly request Stride generation or optimization using existing allowances. Inspect saved results rather than start again.
Work with a local or external model	Discover the full Create setup, request Stride research and brief preparation, retrieve that context, and submit a draft written by your own model. Or save supplied text without new Stride research.
Report work and review outcomes	Record drafts, revisions and genuinely shipped changes; inspect work history and eligible shipment assessments without automatically starting a scan.

You do not need to memorize tool names. Describe the outcome you want, ask your assistant to inspect what is available, and set clear approval boundaries.

How to connect

In your AI app, add an OAuth-capable remote MCP connection. The Stride server address is:

https://app.strideftw.com/api/mcp
Sign in to Stride when prompted.
Check the Stride account and return address, then select the sites you want the assistant to access. Review Requested permissions and expand Full permissions and limits for details before approving.
Ask: “Which Stride sites and tools can you access? Do not change anything yet.”
Confirm the correct site before starting work, especially if you manage several clients.

Connection steps vary by AI app. Read access and revocation have been tested with Claude Code and Cursor. Full Content authoring, work reporting and Actions-management journeys have not yet been confirmed in an external AI app; do not assume every client or onboarding flow has been verified.

You can review and revoke connections in Site setup → Integrations → Stride MCP. After a release, refresh your AI app's tool list, or reconnect if it caches older tools.

Agency client-plan changes do not remove MCP access while the account remains on Agency. If a standalone account moves from Scale to Starter or Grow, MCP tool access stops without deleting its existing grants or saved data. You can still review and revoke the connection. Existing connections do not automatically gain sites or permissions: reconnect and explicitly select any additional sites.

Choose only the permissions you need

Permission	What it allows	What it does not allow on its own
Read	Read approved-site research, inspect Actions and saved work, and view shipment assessments.	Change data, spend generation credits or report external work.
Act	Make supported tracking and knowledge changes, save Content setup or supplied drafts, and explicitly request scans, content generation or Agent work.	Read all research without Read permission, manage Actions without Manage Actions, or publish a page.
Manage Actions	Resume and change supported saved Action work, briefs, draft edits and your personal queue order.	Paid target checks, generation, AI revisions or retries without separate Act permission.
Report work	Begin and update external work records, and confirm that reports were saved.	Access full site research or exposure assessments without Read permission, generate content, or independently verify a shipment.

Permissions do not replace your account's access rules. A view-only site cannot become editable through MCP. Existing connections do not automatically gain new permissions; adding permission requires explicit approval.

For a research-only connection, start with Read. For Stride preparation followed by local writing, you need Read and Act. Add Manage Actions for later saved-work management, and Report work when you want to record activity performed outside Stride.

Useful workflows

1. Understand the site before changing anything

Start with a read-only review:

Review the latest saved evidence for this site. Explain where we stand, which buyer questions need attention, and the three most useful existing Actions. Include the dates and any missing evidence. Do not run a new scan or change anything.

Your assistant can bring together saved visibility, citations, competitors, questions and Actions. It can explain the existing priority rationale and distinguish a fresh finding from older evidence.

Reading saved results does not collect a new scan. Ask for a fresh scan separately if you need one, and check the allowance before approving it.

2. Pick up an existing Action

Avoid creating another draft for work already underway:

Find the existing Action for this objective. Tell me its current stage, what has already been saved, what is blocking it and the next permitted step. Resume that work instead of creating a duplicate. Ask before paid research or generation.

With the right permissions, your assistant can:

Open or resume the existing saved work.
Update supported briefs, page targets and Content setup.
Save supplied draft edits.
Explicitly request eligible generation, AI revision or retry after approval.
Hide an irrelevant queue item or restore a hidden one.
Dismiss eligible saved work, with an explanation.
Reorder your personal To do, In progress or Done segment.
Accept eligible Agent proposals you have reviewed.

Important limits: There is no direct MCP merge or “mark Action complete” command. Hiding and dismissal are not completion or proof. Reopening is limited to a specific recovery case; it is not a general undo for every dismissed or verified Action. Your personal ordering does not change the team's evidence-based priorities.

3. Have Stride generate the content

Ask your assistant to review the setup before spending a credit:

Set up an article for this buyer question. Show the available content types, audience and writing personas, brand rules, Knowledge Base sources and eligible references. Recommend a brief, but wait for my approval before generating.

You can choose the format, length, language, tone, author, audience persona, writing perspective and supported type-specific brief fields. Audience and writing perspective are separate: one describes the reader; the other describes how the piece should sound.

Guided setup: your assistant can use the free, Read-scoped get_content_interview tool to review the answers you have already given, ask only the unresolved material questions and disclose its recommendations. The questions belong in your existing chat; you do not need an installed skill or a client-specific form. Recommendations are not automatically your selections.

Before starting, ask for one compact brief with the buyer question, writer, material choices, constraints and credit cost. Confirm those choices and any spend before the assistant calls the existing execution tools. The interview does not authorize execution or independently prove that you approved; your assistant must follow the approval guidance, and Stride cannot guarantee that every AI client will do so.

A Read-only connection can help plan a brief without being able to execute it. If execution is unavailable, your assistant should explain the missing permission or access rather than ask you to approve a spend it cannot start. Approving a brief does not add connection permissions.

After approval, Stride runs the existing generation workflow. The saved result appears in Content → Create, or the linked existing Action when you started from that objective. Your assistant should inspect the returned saved run rather than create another one because the first response took time. Every Content acceptance or status response also provides an absolute link to the exact Create or Optimize run on the correct site, using the dashboard address even when the MCP service has a separate host.

For an existing page, the assistant can request Stride's existing Conservative, Balanced or Full rewrite optimization. The separate research-only handoff to a local writer described next currently applies to Create, not a new external Optimize engine.

4. Use Stride research, but let your own model write

This is useful when you want Stride's preparation and context but prefer your own writing model.

Review the setup. Choose the buyer question, content type, brief, brand rules, audience, writing perspective, preferred knowledge sources and eligible cited references. When the guided interview is available, it can help resolve missing choices without repeating settled answers.
Approve the brief and cost. Explicit Stride research and brief preparation uses one existing content credit, regardless of which model will write.
Check the same preparation run. A queued request is not ready research or a delivered article. “Not ready yet” is a normal successful pending response, not a failed preparation or a reason to pay again. It contains no preparation text to extract.
Retrieve the saved preparation only when ready. It includes the finalized writing input and the source context actually assembled for that run. Keep it inside the agent/model environment you have approved. Preparation complete is not finished writing.
Write with your chosen model. Your assistant submits the result back to the same prepared run, using the current saved version.
Review in Stride. The accepted draft appears in Content → Create and can be edited there. Submission does not use a second content credit.

Try:

Use Stride to prepare the research and brief for this article, then let my current model write it. Show me the setup and ask before spending the one content credit. Save the finished draft back to the same Stride run. Do not publish it or mark the Action complete.

Knowledge source selections are preferences, not a promise that every source was used or that other eligible knowledge was excluded. Eligible reference choices come from Stride's source picker; the connector is not permission to crawl any URL the agent invents.

Stride labels the result as externally supplied. It does not claim that it wrote, generated metadata for, or ran its readiness review on that draft. A declared model name is reported by the writer, not independently certified. Linking the preparation does not prove that the model followed every instruction or used every source.

Waiting and resuming: by default, your assistant should return the pending status and exact-run review link promptly. If you ask it to wait, it can check that same run with get_content_run at the suggested 15-second interval, for no more than four checks or 60 seconds, whichever comes first. This is an optional bounded loop in your assistant, not a request held open by Stride or a promise of a later notification. If work is still pending, the assistant should hand back the link and explain the remaining stage, then resume the same run when you return.

A delayed response must not automatically start another preparation or generation attempt, publish content or complete an Action. For a retry of the identical request, the assistant should reuse the original request key; a status check uses the existing run ID.

5. Save a draft you already have

You do not need Stride preparation to save supplied text:

Save this supplied draft in Content → Create for this site. Keep it as a draft. Do not request new research or generation, add tracked questions, publish it or complete an Action.

Saving supplied text without new Stride preparation uses no content credit. It still needs Act permission and current edit access. If you later manage that accepted draft through the Actions connector, the supported saved-draft editing operation requires Manage Actions permission.

Review the title, metadata, body and formatting in Stride. Saving source or persona selections alongside the draft is not evidence that Stride researched those sources or that your model used them.

The browser editor is not a general-purpose Markdown or HTML editor. Keep your original file and check specialized formatting before editing in the browser. If exact formatting matters, use your own writing tool and have the assistant save against the latest draft version rather than overwrite newer edits.

6. Keep outside work visible in Stride

Use Report work when the substantive work happens outside Stride, such as local writing or website code changes.

Begin a work record for this change before we start. Link it to the existing Action where appropriate, record the buyer questions and signal we want to watch, and keep drafts and revisions on the same work item. Report it as shipped only after the change is actually live. Confirm that each report was saved.

Your assistant should:

Begin one work record before substantial work. If you want later measurement, identify the existing buyer questions and the signal to watch before making the change.
Record a produced draft and subsequent revisions as different stages of that same work.
Report a shipment only after it really went live, with the correct public target and shipment time.
Read the saved record back to confirm that reporting succeeded.

You do not need to install a reporting skill. The MCP connection supplies reporting instructions. Your agent still has to follow those instructions and report the result; Stride cannot watch arbitrary files or work happening on your computer.

Supported work performed through Stride can be linked automatically when work reporting is enabled. That does not make every read or local task an exhaustive audit record. If the agent produced work but could not save its report, the honest result is “Work produced; reporting not saved.”

In Work & contribution, the overview follows your selected period and work-history filters. History rows describe the operation and its Action, such as “Updated content setup: Improve the buying guide,” when that context is available. Open a row for evidence, the linked Action, client-export selection and the name of the user who authorized the connection. An unavailable name is labeled rather than replaced with an email address.

Older entries can recover an operation from its exact saved receipt. Any recovered Action title is labeled as current saved context, not a historical snapshot. Missing receipts or Action links are not guessed. A completed operation is still not proof of publication, independent verification or improved exposure.

7. See what happened after a change shipped

Ask for an assessment, not a claim of guaranteed impact:

Show the shipment assessment for this change. Separate what we reported, what Stride independently verified and what the exposure readings show. Explain the before-and-after evidence, any overlapping changes, and why a comparison may still be unavailable. Do not start a paid scan.

The Work & contribution → What changed after shipment view and the corresponding read tools use the existing shipment and measurement history.

Open a shipment row for its before-and-after observations, proof limits and next checkpoint. Shipment totals cover retained history for the selected shipment origin, independently of the work-history period above.

A useful comparison needs an honest starting point and enough comparable later observations. Stride may show that it is waiting, that evidence is insufficient, or that other changes make interpretation difficult. A missing baseline is unknown, not zero.

The current directional page-content assessment is not a dedicated calculator for every technical change or off-site placement. An unsupported comparison should remain unavailable rather than borrow unrelated results.

Improvement after a change is an association, not proof that the change caused it. Independent live verification, exposure changes and financial ROI are different claims. Reading an assessment does not automatically verify a page, start scans, publish content or complete an Action.

Review selected items and their client-visible references before exporting or sharing a report. Nothing should be shared merely because an agent says the work is finished.

Credits and limits

MCP uses the existing Stride allowances; it does not introduce a separate generation-credit pool. Your AI app or external model may have its own charges.

Activity	Stride allowance or cost
Read saved research, inspect Actions, review a Content interview when available, or read shipment assessments	No scan, content or Agent credit for the read itself.
Report work or confirm a saved report	No scan, content or Agent credit.
Save setup or import supplied text without new research	No content credit.
Request Stride research and brief preparation for external writing	One existing content credit after explicit approval; submission uses no second credit.
Request Stride content generation or a Stride AI revision	One existing content credit per approved generation or Stride AI revision, subject to the existing allowance and failure/retry rules. Editing supplied text with your own model does not itself spend a Stride content credit.
Request a scan	The site's free first scan or an existing scan credit afterward, where eligible.
Ask Stride's Agent to perform work	Existing Agent-run allowance applies.
Request target checks or other supported changes	Existing access, research, rate and plan limits apply; not every check consumes a content credit.

Opening saved work is not the same as restarting it. A generation retry can create a new attempt and use the normal allowance. Ask your assistant to explain the cost before approving one.

What the status actually means

Status or claim	Meaning
Requested or queued	Stride received the request or accepted it for processing. The result is not finished yet.
Preparing or running	Work is underway. Do not start a duplicate merely because it takes time.
Preparation not ready yet	A successful pending read, not a failed operation. No preparation text is available yet; check the same run later.
Prepared / waiting for external writing	Research and brief preparation finished, but an external draft has not been delivered.
Draft / accepted external draft	Text is saved for review. It is not published, independently verified or automatically marked complete.
Completed operation	The requested operation finished. Completing preparation or saving a draft does not mean a site change shipped.
Reported shipped	The agent reports that the change went live. This alone is not independent proof.
Independently verified	A separate supported check passed. Its scope and date matter; it does not prove authorship, quality or causal impact.
Measuring / waiting for evidence	Stride is assessing eligible observations, or explaining which prerequisite is missing.
Failed, cancelled or unknown	The operation failed, stopped, or has an unresolved outcome. Do not treat it as successful or blindly start it again.

Privacy and control

Approve only the sites and permissions the assistant needs. Revocation or losing your site access stops future authorized use; revoking a connection does not delete your saved Content or all historical shipment records.
Prepared writing context can contain private source excerpts. Use only an AI app/model environment you trust with that material.
Save complete drafts through the dedicated Content workflow. Work reports should contain safe summaries and references, not private drafts, source packets, passwords, API keys or arbitrary local file contents.
Public report references must be safe public HTTPS links. Private or signed download links are not appropriate public evidence.
Private preparation packets expire after 30 days. Your saved draft is not deleted because its preparation packet expires.
Detailed work reports have a 90-day retention window. That is different from the retained shipment/implementation history; revisions do not restart the detail-retention clock.
Review exports before sharing. Protected fields are excluded, but you are still responsible for the titles, summaries and references you choose to disclose.

FAQ and troubleshooting

Why can my assistant read the site but not edit it?

Read permission does not include Act or Manage Actions. Your account and site role must also allow editing, and the requested feature must be operational and within your plan allowances.

Why are some tools missing?

Check the selected sites, approved permissions and current account/site access. The service or individual capabilities may also be temporarily disabled by operational controls. Refresh tool discovery after a release. Reconnect to add a missing permission; refreshing a token does not silently add it. Installing a skill does not unlock tools.

Do I need both Act and Report work to save an external draft?

No. Saving a supplied Content draft needs Act and edit access; the full research handoff additionally needs Read. Report work is separate and optional for this Content journey. Use it when you also want cooperative reporting of work done outside Stride.

Why does Stride say my local draft was not reviewed?

Because your model supplied it. Saving it in Stride, or associating it with Stride research, is not the same as Stride running its own writing, metadata or readiness steps.

Can the agent publish the page or mark the Action complete?

Not through this Stride MCP service. Publish through your normal CMS process and use Stride's separate publication/proof workflow afterward. The separate CMS connector has its own permissions and availability; this connection does not inherit them.

The request timed out. Should I ask the agent to start again?

Ask it to inspect the saved run or work receipt first. It should retry the same intended request with the same request key, not create another paid run or a parallel Action to repair a missing response. It should retain the exact-run review link and use the bounded same-run checking policy above rather than wait indefinitely. If the draft changed elsewhere, reload the current version before saving over it.

Why is my work missing from the statistics?

Drafts and revisions can appear in work history without counting as shipped changes. Repeated saves are not separate delivered articles. Outside work must be reported, and exposure assessment needs eligible scope, proof and comparable observations. Unreported work is not evidence that no work was done. Missing data should be explained, not filled with invented results.

Does using the connector guarantee better visibility?

No. It helps your assistant use the right context, continue the right work and keep honest records. Visibility outcomes still depend on what changes, what actually ships, and the evidence collected afterward.

A safe first session

Start with this:

List the Stride sites and features I have approved. For the site I choose, review saved visibility and existing Actions, then recommend one next step with its evidence and cost. Resume existing work where possible. Ask before paid work. Do not publish, mark Actions complete or report a shipment unless it really happened. Confirm any saved work before telling me it is synchronized.

Then try a read-only review or a no-research supplied-draft import before approving paid preparation or generation.

Integrations
Actions and proof
Agent activity — site visits by crawlers and agents, not the same thing as MCP work reporting.