Run the engines from your AI assistant
Torke Engine is a set of tools for campaigns on approved claims, account signals, inbound response, first-touch outreach and a weekly performance read, plus the Torke MCP connector that plugs them into Claude, Codex or any AI assistant that supports MCP. Your people ask in plain words, the assistant drafts and checks, and nothing reaches a buyer until a signed-in person on your team releases it.
What is Torke Engine?
Torke Engine connects campaign, signal, inbound and first-touch tools to Claude, Codex and other MCP assistants. The Reachout Engine checks evidence for cold email drafts and does not send them. An optional referral step is planned: it could use client-supplied relationship records or a team member's own LinkedIn check to flag a connector, draft a separate request and record the response. A person verifies, reviews and sends it. This step is not in the running Engine.
Why run marketing from an AI assistant now?
Each figure links to the page it comes from, so you can check it before you repeat it.
| What the evidence says | Source |
|---|---|
| Anthropic released the Model Context Protocol as an open standard for connecting AI assistants to the systems where data lives. | Anthropic (25 November 2024) |
| New GTM engineering job postings grew 205% year over year in 2025. | Bloomberry job postings analysis (2025) |
| 45% of B2B buyers reported using AI during a recent purchase, and 67% prefer a rep-free buying experience (survey of 646 buyers). | Gartner (9 March 2026) |
How the assistant does the work
MCP, the Model Context Protocol, is an open standard that lets an AI assistant call outside tools. The Torke MCP connector makes the Engine's tools available to the assistant your team already uses.
-
Ask in plain words
A marketer asks for three search ads for next month's warehouse automation campaign, and the assistant calls Torke to write them and return files ready to import.
-
One record behind every answer
Your approved facts, word rules, accounts and campaigns sit in one place, so every request draws on the same source and nobody keeps a second login.
-
A person releases
The assistant can draft, check and propose. Anything that would leave the company, an ad, an email or a page, waits for a signed-in person to approve it.
What each module produces
Each module states what changes for your team, how it works and where it stops, because the limit is what your legal and technical colleagues will ask about first.
-
Campaigns on approved claims
Ads for Google search, LinkedIn and Meta, newsletters and web copy where every line matches what your company approved.
How it worksEach line is checked against your approved facts and word rules before it can be exported, and you receive files in each platform's own import format, with paid campaigns paused.
Where it stopsWe never launch spend. A person on your team imports and starts every paid campaign.
-
Account signals
Each morning, the companies that read your site and the pages they read, plus dated public events at your target accounts, ranked, with the opening line the strongest event justifies.
How it worksIt uses your own site data and free public sources, such as newsrooms, job boards, watched pages and regulator feeds, and older events count for less each week.
Where it stopsNo LinkedIn scraping, no data brokers and no fingerprinting. Accounts with no public footprint rank on fit alone.
-
Inbound response
Every inbound request qualified, routed to the right rep by language and segment, and answered in the buyer's language with meeting times that are free on the rep's calendar.
How it worksWritten rules score each lead, and the calendar is read again just before a time is offered. A lead that asks to talk now is answered within minutes, with escalation when a deadline passes.
Where it stopsIt can run in draft mode, where a rep sends each reply, and it never offers a time it hasn't checked again.
-
First-touch outreach
First emails to named people where every sentence traces back to something that person or their company said.
How it worksThe research behind each email is checked before writing starts, and anything inferred has to be phrased as a question.
Where it stopsIt never sends. A person reads and sends every email.
-
Performance read
A weekly read of what each campaign produced, judged against the target set when it started.
How it worksA change is called only when the sample is large enough to support it, and each lesson stays proposed until someone on your team confirms it.
Where it stopsA week with too little data is reported as unreadable, never smoothed into a trend.
-
One rule across all five
The Engine proposes and checks, and your people decide. Every module ends at the same point, a signed-in person on your team releasing the work or sending it themselves.
Read the commitments
Planned: optional referral step
The Reachout Engine checks evidence for cold email drafts and does not send them. We are planning an optional branch that flags a possible connector from relationship records your team provides or a team member's own manual LinkedIn check, prepares a separate introduction request, and records whether the connector agrees, declines or does not respond. A person would confirm the relationship, review and send the request. This branch is not yet part of the running Engine.
Check the possible path. Use a client-supplied CRM or contact record, or a team member's direct check. Store the evidence reference, verifier and date without scraping LinkedIn or guessing how close the relationship is.
Draft a separate request. The assistant prepares a short note to the possible connector from approved information. It does not contact the prospect or send a LinkedIn message.
Leave the decision with a person. The employee confirms the relationship, reviews and sends the request, then records the response. An agreed introduction does not replace the normal checks before contacting the prospect.
The Torke Nodes inside the Engine
Torke Engine is built from seven nodes, each handling one stage of the work. You can rent or buy one node on its own and add the others later, and each node page states its price.
-
Context
Torke Context
Your approved facts, word rules and examples loaded for every AI request, and each draft checked against them before anyone reads it.
How Torke Context works -
Signal
Torke Signal
Watches the companies you sell to every morning, ranks them on dated public events and emails you the ones worth acting on, with the proof.
How Torke Signal works -
Signal
Torke Visitor Signal
A morning email naming the companies that read your website, which pages and for how long, with no visitor IP address stored.
How Torke Visitor Signal works -
Plan
Torke Router
Scores every campaign step and routes it to code, a model, a hybrid or a person, with its test and fallback, before anything is built.
How Torke Router works -
Outbound
Torke Reachout
A first email to one named buyer, built on their own public words, with every sentence checked against its source before a person sends it.
How Torke Reachout works -
Inbound
Torke Qualification
Reads each inbound lead, scores it in code and replies in the lead's language with meeting times checked against the rep's live calendar.
How Torke Qualification works -
Measure
Torke Ledger
Google Ads, Search Console and GA4 in one file, with a weekly review that judges each campaign against its own target.
How Torke Ledger works
What your team gains
We'll publish measured figures once a client account has produced them. Until then, this is what the Engine is built to change.
Fewer hours from brief to launch
Drafting, claim checking and platform formatting happen in one request, so the hours go to deciding what to say.
No unapproved claim in front of a buyer
A line that doesn't match your approved facts can't be exported, which is what a technical evaluator would catch first.
Replies in minutes
Inbound requests are qualified and answered with checked meeting times while the buyer is still at their desk.
A plan built from your own results
Next quarter's plan starts from lessons your team confirmed, rather than from a benchmark taken from someone else's market.
Combine the Engine and the service
Torke Service runs on the same Engine, so moving between them keeps your record intact.
Service first, Engine later. We run the work while the rates are proven, then hand the Engine and its operating record to your team when you hire for it.
Engine first, service for one practice. Your team runs campaigns on the Engine and brings us in for the one practice it can't staff, such as Demand.
Fix, then either. The Diagnostic finds one barrier, we fix it, and you choose who runs what comes next.
How the Engine is priced
Renting all seven nodes together costs CHF 5’900 a month, against CHF 7’870 for the same seven rented one by one. A single node starts at CHF 690 a month, and each node page lists its three options. Prices exclude 8.1 percent VAT, and the pricing page covers Torke Service.
What do teams ask about Torke Engine?
Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is.
What is MCP in marketing?
MCP, the Model Context Protocol, is an open standard Anthropic released on 25 November 2024 that lets an AI assistant call outside tools. In marketing it lets an assistant such as Claude read campaign data and run tools such as ad drafting or account research.
Can Claude create Google Ads campaigns through MCP?
Yes, with a connector. Torke Engine returns search ads in Google's import format with the campaigns paused, after checking every line against the facts your company approved, and a person on your team imports and starts each campaign.
Is it safe to let an AI assistant run our marketing?
Only with a release step. Torke Engine lets the assistant draft, check and propose, while every ad, email or page waits for a signed-in person to approve it, and it never starts spend or sends email on its own.
How do we find account signals without scraping LinkedIn?
Use first-party site data and free public sources such as newsrooms, job boards, watched pages and regulator feeds, each event stored with its date and link. Torke Engine's account signals module works this way, with no data brokers and no device fingerprinting.
Bring us the problem, and we bring the solution to it
Tell us what your team sells, who buys it and where deals stop moving. You'll get a written reply saying which way of working fits you, or whether a Diagnostic makes sense, and if neither does, the reason.