Right now, asking Claude or ChatGPT about a Leadstore account gets an answer. The next step should get a result. Here is what that looks like, and why it is not live yet.
What exists today
The connector has seven tools, and every one of them reads: visibility score, competitors, content calendar, blog drafts, mentions, share of voice, connected sites. A question gets a real number back. An assistant cannot yet be asked to do something and have it happen.
What action tools would let you do
■
Kick off a visibility audit without opening the dashboard first.
■
Generate a blog draft from inside the same conversation where the need came up.
■
Publish a draft that has already been reviewed and approved.
■
Add or remove a tracked competitor.
Why it's not built yet
The authentication side is already solved. Every connection runs through a user's own Leadstore login, per user, and that same login would work for a write action exactly as it works for a read today. What is not solved is the judgment layer sitting in front of it: which actions are safe to let an assistant take with a single request, and which need an explicit confirmation step first. Publishing a blog post to a live site is the clearest case for a confirmation step. Running an audit, which does not change anything customer-facing, is a much easier call.
What the read tools already got right
The seven tools that exist today all share one property worth keeping for whatever comes next: each one does exactly one specific thing, with a name that says what it does. There is no single do-anything-with-my-account tool, even for reads. A tool called publish this draft is a very different thing to grant than a tool called make changes to my account, even if the underlying access looks similar from the server's side. Specific, named actions are easier to reason about, easier to confirm before running, and easier to explain in a permissions screen.
What a mistake would actually look like
Say an action tool to add a tracked competitor existed today, and someone asked an assistant to add our biggest rival in the enterprise segment. A model can misread which company that means, especially for a brand with a name close to another company's. A read tool getting that wrong just shows a wrong name that can be dismissed. A write tool getting it wrong adds a real row to a tracked list, one that might not get noticed until it is already shaping a competitive analysis. That gap, between a wrong answer that can be ignored and a wrong action that has to be noticed and undone, is the reason this is moving slowly.
What a first version might look like
A reasonable first pass keeps the pattern from the read tools: narrow, specific actions rather than one open-ended tool. Kick off an audit. Generate a draft. Nothing that touches a live site without a separate approval, at least at first. That is a judgment call still being made, not a shipped decision, and it will likely stay conservative longer than feels necessary, on purpose.
How this fits into the three Leadstore stages
■
Stage 1, account + site added: users get the audit reward before any tag or key - four core channel scores, top fixes, readiness, generated llms.txt and llm-data.json, and the Leadstore Tag snippet to install.
■
Stage 2, tag installed: users get the live telemetry reward with no key required - AI-crawler activity by engine, verified-bot counts, human traffic, top sources, top pages, devices and geography.
■
Stage 3, key connected: users add answer tracking on top of the audit and telemetry layers - Visibility Score, SAIV, Brand Monitor, share of voice, citations, competitors and hallucination checks.
Next step
Try the read-only MCP connector today, and share what action you would actually reach for first.
