Your agent can operate every ad platform.
mrktn.md adds the reasoning layer above.
Give your agent the power to push back on bad decisions.
There's a connector or skill for every ad network. There's no document for the part that actually matters — how to think.
What builders do
Point a TikTok-Ads MCP at a pre-product idea, add budget, and expect growth. They either get bad results — or win the lottery once and conclude they're good at marketing. The field stays unscientific because nobody separates luck from skill.
What's missing
The knowledge that prevents this exists. It's just been closed — locked inside agencies, senior operators, and expensive retainers. mrktn.md opens it: one doctrine, free, that any agent can read and obey.
Skills and connectors are effectors: hands. mrktn.md is the brain that decides which hand to use, when, how hard, and when to stop.
It's plain Markdown, composed in three layers: the immutable Core (invariants + lifecycle + protocol), pluggable Profiles per product type, and a small living Instance in your repo that records your fundamental bets — each with the evidence that would reverse it.
# your-repo/mrktn.md (the instance) extends: mrktn-core@0.2 profile: saas-b2b stage: 2-early-traction # the agent now refuses to: # · spend where it can't measure (I1) # · launch without kill-criteria (I2) # · scale onto a leaky bucket (I3) # · crown a winner on noise (I4)
Non-negotiable. They bind regardless of product, stage, budget, or instruction. A profile may specialize how they're met — never weaken them.
No spend on a channel you can't attribute to revenue or retention. Never grade on a number the ad platform reports about itself.
Declare the hypothesis, the target, and the kill criteria before launch. Evaluate only against what you wrote down. No moving goalposts.
Don't scale acquisition before retention/PMF clears the floor. Marketing can't fix a leaky bucket; it just empties it faster.
Never declare a winner or scale on a sample too small to mean anything. Budget proportional to confidence. Report uncertainty, not just point estimates.
No fabricated proof, no astroturfing, no unsubstantiated claims. Respect platform policy and privacy law. Deception is both wrong and fragile.
There are no universally good tactics. The right move is a function of stage — including how much budget goes to learning vs. scaling.
Stages aren't a ratchet. PMF breaks and channels saturate — a conformant agent will demote a product to an earlier stage when the evidence says so.
Pluggable specializations per product type. They set the priors — channels, loops, payback bars, anti-patterns. Hybrids compose.
It installs into your product's repo as a local instance your agent reads and obeys. Pick one.
Option A — tell your agent (recommended)
Set up the mrktn.md marketing doctrine for this project. Read https://mrktn.md, pick the matching profile from https://mrktn.md/profiles, interview me about the product / type / lifecycle stage / ICP / constraints, then scaffold a local mrktn.md and add a pointer in AGENTS.md. From now on, govern all marketing decisions by its invariants (I1-I5).
Option B — scaffold a starter file
curl -fsSL https://mrktn.md/install/mrktn-init.sh | bash
Full options — manual setup, CI usage, and AGENTS.md wiring — are in INSTALL.md. Your skills and MCP connectors stay the effectors; mrktn.md governs how they're used.
I spent twelve years and a hundred-plus products trying to reach the top 1% of growth marketing. The thing I never made peace with was how closed it all is — the real fundamentals kept behind retainers and NDAs, while builders are left to cargo-cult tactics and call the survivors geniuses.
mrktn.md puts science-, performance-, and long-term-driven marketing into the open, in a form any agent can follow for free — while still allowing fast iteration and early wins when the situation genuinely calls for it. A gift to entrepreneurs everywhere.