MCP playbooks
Recorded sessions on the Linkeddit connector
A playbook here is a real transcript: the ask, the tool call with its arguments, and the response as it came back, abridged only by omission, with the account, project and date it was recorded on. No mock output. If a step was not executed, the page says so and quotes the tool's contract instead.
Playbooks
An AI visibility audit, run from Cursor
Seven read calls: projects, configuration, credits, the last run, per-engine metrics with their disclosure, and the ranked gaps. Recorded 4 September 2026.
7 calls, reads onlyRead VS CodeThe fix-and-verify loop, from VS Code
Open the top gap with its citations, read the fan-out, check a draft against the recorded evidence and watch the guard reject an uncited source, then register a re-measurement.
3 live calls, 3 contractsRead Any clientGrounding a prompt set in Search Console
The honest not-connected response, the two-step service-account connection, and the comparison your client runs against the approved prompt set once it is.
2 live calls, 1 connectionRead Claude DesktopBuyer-intent search, from Claude Desktop
Search Reddit, refine a noisy query, read the thread where someone tested ten AI visibility tools, and turn its objections into prompts to measure. Two thin results kept as recorded.
4 live calls, 1 contractReadHow a playbook is recorded
Four rules, so a transcript here can be trusted as evidence rather than read as marketing.
Real calls, real responses
Every response block is copied from the live tool and cut only by leaving fields out. Numbers, ids, engine names and instrument tokens are as returned. Where a call was not executed, the block is labelled as a contract quoted from the tool description.
Our own data, or a test account
Recordings run against Linkeddit’s own linkeddit.com project, or a standing test account. Never a customer’s. That is why two Compete playbooks are listed as planned below rather than published.
Reads by default
A recording should not leave artifacts behind to look complete. Write steps that create rows are quoted, not run, unless the page says otherwise.
Dated, and refusals kept
Each page states its recording date. When a tool refuses, as gsc_queries_get does for an unconnected project, the refusal stays in the transcript, because the refusal is the behaviour a buyer should see before paying.
Planned, not yet recorded
Named here so a title is never mistaken for a shipped transcript.
- Weekly competitor brief from Claude Code (Compete): waiting on a test account, so the recording does not publish a real customer's tracked competitors.
- Competitor complaint sweep from ChatGPT (Compete): same dependency.
Related
Connector documentation
Quickstart, authentication, client guides, the tool reference, rate limits and troubleshooting.
Read HelpAnswer Radar from MCP clients
Which tools exist, how scopes and the Compete plan apply, and a first workflow.
Read MethodologyHow the numbers in these transcripts are defined
Six metrics, admissibility, segments, versions, the faithfulness guard and re-measurement.
ReadConnect and run one
The connector uses OAuth 2.1; there are no keys to paste. Answer Radar tools require the Compete plan.