Bring RunPR research into your AI assistant

You may already have a research checklist or pitching skill you use with an AI assistant. You know how to ask for a useful first draft. Bringing that process into client work means giving the assistant the right brief, checking its sources and keeping the draft somewhere your team can review it.

RunPR’s limited-access OAuth pilot connects those steps. Through MCP, the Model Context Protocol, an enrolled user can give Claude Code access to selected client context and saved AI Visibility evidence, research a relevant byline and submit a draft to RunPR for review.

Access requires enrollment. Ask for pilot access so I can enroll you and help you start with one client in Claude Code. The pilot supports Claude Code, tested on macOS; other assistants remain unverified. Existing support-provisioned organization-key connections remain available separately; they do not have OAuth’s client-selection controls.

Keep the client context attached to the research

A useful pitch needs a contribution the client can support and a reason a particular reporter would care. Start by asking your assistant to retrieve the selected client’s saved facts, expertise and voice. Then inspect a completed AI Visibility scan, keeping the question, date and cited pages alongside the answer.

The question matters. A source cited in a product comparison may serve a different purpose from one cited in an answer about an industry problem. Reading the saved evidence does not start a new scan, and the assistant should tell you when the information is dated or incomplete.

Investigate the citation before choosing a reporter

An engine may cite a page because it explains a category, lists products or summarizes somebody else’s work. That does not establish that its author covers your client’s subject or wants a pitch. Citation frequency also does not tell you whether a placement will cause future AI mentions.

Use the citation as a lead: inspect the answer, open a relevant source, check the article and byline, then decide whether your client has something useful to add. If the byline is missing or the client’s evidence is too thin, ask the assistant to flag the gap.

For example:

For the selected client, inspect the sources from the latest completed category-discovery scan. Pick one relevant article, identify its author if available and explain what our client could contribute. Prepare a short pitch with source links and a list of facts we still need to verify.

This is an illustrative request, not a reported customer result. You might end up with a draft, a request for a missing fact or a decision to leave the article alone.

Bring your own writing process

Your assistant can use the prompts or skills it already understands to organize research and compose the pitch. RunPR provides permitted data and tools; it does not import or run those skills for you.

After you have a draft, ask the assistant to submit its text to RunPR. RunPR performs a bounded draft review and saves it for your team to inspect, including any hold reason. It does not require a second generation step to rewrite the assistant’s draft. Research and review still use RunPR resources and allowances, with additional pilot limits.

Open the review link your assistant returns, check the supporting evidence and make your edits in RunPR. If RunPR holds the draft, inspect the hold reasons before proceeding. Your team decides whether to approve it and what to send. The OAuth tools cannot approve, send or delete work.

Choose a client and a defined set of permissions

After enrollment, follow the Claude Code commands in the setup guide to connect and sign in. Ask it to call connection_info, then open the setup link it returns. Choose your RunPR workspace, select the clients the assistant may access and save the permissions you want to grant. Return to the assistant to continue. The setup guide explains the connection steps and each permission.

RunPR checks the selection when a tool runs. An instruction to fetch another client’s brief cannot grant access to it. Reading existing evidence, researching a source and submitting a draft require different permissions; you can keep a connection read-only.

RunPR keeps one active selection for each user and OAuth client. Disconnecting in RunPR stops that connection’s access. If you later reconnect explicitly, still-valid tokens for the same assistant can use the new selection within their permissions. Remove the assistant’s provider authorization too when you retire it. These controls do not erase information the assistant has already retrieved, so check its data policies before connecting.

Try one research-to-review handoff

Start with a client whose brief and completed scan you can check yourself. Ask the assistant to explain one cited source, identify the relevant author if available and draft an approach based on a fact your client can substantiate. Then review the result in RunPR.

The Claude Code pilot requires explicit enrollment; it is not available automatically to every workspace or MCP client. Explore RunPR MCP for the access model, or ask for pilot access with the assistant and client workflow you want to use.