Docs/Agent Skills

Agent skills

Drop-in skills that teach your agent to run Coppica: generate copy, track what converts, and run the optimization loop. There is nothing to install or host, so these aren't a server. They're the playbooks.

The single highest-leverage step. Paste this into your agent's main memory file (CLAUDE.md, AGENTS.md, a Cursor rule, or your system prompt) so it uses Coppica proactively, not just when told. Without it, an agent connects and then forgets the tools exist.

## Coppica (persistent reminder)

You have Coppica connected via MCP. Reach for it proactively whenever a task touches a
client's marketing, and prefer its tools over generating or reasoning from scratch.

- Discussing strategy, positioning, offers, audience, or messaging -> first pull the
  client's context from Coppica (coppica_get_client_context, intake, avatars, offers,
  strategic docs). Reason from their real strategy, not generic assumptions.
- Asked to write or improve marketing copy (ads, emails, pages, hooks, scripts) ->
  generate it with Coppica (coppica_write_copy, or coppica_generate_ad_set for a set)
  instead of writing raw. The output is grounded in proven patterns and is trackable.
- Copy gets deployed anywhere -> close the loop: attach a tracking link and report
  conversions with coppica_report_outcome.
- Improving results over time -> run the optimization loop (rank, diagnose, iterate,
  redeploy with tracking) on a cadence.

Default: if the work involves a client Coppica knows about, use Coppica's tools before
falling back to your own generation.

Download coppica-memory.md for the full version with per-tool placement notes.

Get a cpk_ key from Coppica → Settings → API Keys, then point any MCP-capable agent at the hosted endpoint. It exposes Coppica's full tool set; every request uses your own key.

{
  "mcpServers": {
    "coppica": {
      "type": "http",
      "url": "https://mcp.coppica.com/mcp",
      "headers": { "Authorization": "Bearer cpk_YOUR_KEY" }
    }
  }
}

REST is available too at https://coppica.com/api/v1 with the same header. The cpk_ API key is separate from the ck_ website pixel key.

Claude Code / Agent SDK

Drop the skill folders into your project (or ~/.claude/skills/). Each is a SKILL.md your agent auto-loads at the right moment. Then add the MCP config so the agent can call Coppica.

Nine small skills. Your agent reads Overview first; it routes to the rest as the task demands. The loop skills are where the compounding value lives.

Entry

Overview

Object model, the golden rule, and which skill to load for each task. Read first.

View raw →

Connect

Connect

Get a cpk_ key and connect via the zero-install hosted MCP endpoint.

View raw →

Core job

Onboard a client

Create a client and populate its intake before generating.

View raw →

Core job

Generate copy

Choose a bot, generate, and finalize a draft.

View raw →

Core job

Generate an ad set

Server-orchestrated angle to hook to ad set, with lineage.

View raw →

Loop

Close the loop

Track every deploy, report every conversion. The non-negotiable.

View raw →

Loop · Flagship

Optimization loop

The flagship recurring routine: rank, diagnose, iterate, redeploy, report.

View raw →

Loop

Backfill existing copy

Import copy you already have so the taxonomy starts warm.

View raw →

Reporting

Signal report

Turn attribution data into a plain-language recommendation.

View raw →

Most AI copy tools stop at generation. Coppica's skills encode the whole loop: generate copy, deploy it with a tracking link, report outcomes when conversions happen, and let the taxonomy reweight from that real signal, so the next generation is informed by what actually worked. The optimization loop skill runs that on a cadence: rank winners, diagnose why they won, iterate on them, redeploy with tracking, and report. Run it weekly and the small improvements compound.

For field-level detail on every endpoint, see the API reference; for narrative walkthroughs, the guides.

← Back to docs