Bring your PR research into the assistant you use.
Keep your own prompts and writing workflow. Use RunPR for the client context, source evidence and review queue that make the work useful to your team.
Limited-access OAuth pilot for Claude Code, tested on macOS. Ask RunPR to enroll you before setup. Existing organization-key connections remain separate and do not provide OAuth’s client-selection controls.
From a cited source to a draft worth reviewing.
Start with a completed AI Visibility scan, investigate a cited source and bring your draft back to RunPR for human review.
Start with the client
Give your assistant the client facts, expertise and voice you have saved in RunPR. Keep the brief tied to the client you select.
Inspect the evidence
Follow saved AI Visibility answers back to their questions, dates and cited pages. A citation is a research lead, not proof that an outlet or reporter is a good fit.
Research a possible approach
Check the relevant article and byline, then connect the reporter’s work to something your client can credibly contribute. Leave unsupported angles out.
Return a draft for review
Use your assistant’s writing process and save the resulting pitch in RunPR. Open the returned RunPR review link to check the evidence, any hold reason and the draft before deciding what happens next.
Ask for the reasoning behind the pitch.
For the client I selected, show the sources cited in the latest completed category-discovery scan. Inspect one relevant article and its byline, explain what the client could add, and prepare a draft for review. Include the source links and flag anything we still need to verify.
This is an illustrative request, not a customer result. The quality of the answer depends on your saved evidence, the client’s documented expertise and your assistant’s support for the available tools.
Choose what your assistant can work with.
Choose one workspace, select the clients your assistant can access and grant only the permissions it needs.
Read existing information
Retrieve permitted client context and saved evidence. Reading a completed scan does not start another scan. Read access does not grant research or draft-writing actions.
Research and prepare
Grant research and draft submission separately from read access. Your assistant supplies the draft text; RunPR checks it for review. Research and review use RunPR resources and allowances. Pilot limits and your workspace’s billing and product access still apply.
Review stays with your team
The MCP integration has no approval, sending or deletion tool. A submitted draft must go through RunPR’s review workflow. Your assistant cannot send it by asking for a broader action.
Understand where data goes
Your chosen assistant receives the information you permit it to retrieve under that service’s data policies. Disconnect in RunPR to stop the connection’s access, and remove the assistant’s provider authorization when retiring it. Neither action erases data the assistant already received.
Set up your assistant and choose its access.
MCP, the Model Context Protocol, lets an assistant call a defined set of RunPR tools. The OAuth pilot uses Streamable HTTP at https://runpr.ai/api/mcp/oauth. The pilot supports Claude Code, tested on macOS. Other assistants remain unverified. Follow the Claude Code commands in the setup guide after RunPR confirms your enrollment.
Connect from Claude Code and sign in to authorize it. Ask it to call connection_info, then open the returned setup link. Select your RunPR workspace, clients and permissions, save the connection, and return to the assistant. Opening Settings by itself does not start a new connection.
A skill or prompt you already use can guide your assistant’s research and writing. RunPR does not import or execute that skill for you, and an assistant’s instructions cannot expand its RunPR permissions.
RunPR keeps one active selection per user and OAuth client. Reconnecting explicitly replaces that selection; still-valid tokens for that assistant can use the new selection within their permissions. See the OAuth setup guide for scopes, limits and disconnect behavior.
Existing support-provisioned organization keys use https://runpr.ai/api/mcp. They have organization-wide access, are not service tokens and must not be used on the OAuth endpoint.