# Attribuly MCP > Use Attribuly MCP data and a marketing encyclopedia in Codex for ecommerce attribution, CMO-style diagnosis, campaign performance, ROAS/CPA/MER analysis, platform-vs-Attribuly discrepancy checks, product analytics, influencer attribution, tracking-link reporting, paid ads optimization, creative iteration, CRO, SEO/AEO/GEO, content operations, lifecycle/retention, outbound, sales pipeline, revenue intelligence, offer/pricing diagnosis, and AARRR growth planning. Trigger when the user asks Codex to query or analyze Attribuly, connect to Attribuly MCP, build marketing performance reports, diagnose Meta/Google attribution issues, optimize budget/targeting/creative, plan experiments, create channel/content/SEO/outbound strategy, explain what to do next from data, or summarize DTC ecommerce growth insights. - Skill: `zhoushunke0613/attribuly-mcp` (Agent Skill, multi-file: 7 files) - Install (CLI): `npx skillmds@latest add zhoushunke0613/attribuly-mcp` - Raw SKILL.md: https://api.skillmd.com/api/skills/zhoushunke0613/attribuly-mcp/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: zhoushunke0613 (https://skillmd.com/u/zhoushunke0613) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/zhoushunke0613/attribuly-mcp --- # Attribuly MCP Use the Attribuly MCP connector to answer attribution and ecommerce marketing questions from first-party Attribuly data. Prefer the MCP tools over manual API calls whenever the connector is available. ## Connection Expect an MCP server named `attribuly` or another clearly labeled Attribuly connector. The current hosted server is `https://data.api.attribuly.com/mcp` and authentication is configured with an `ApiKey` header in the connector or host configuration. Do not ask the user to paste API keys into chat. Do not print, log, echo, or summarize any key, header value, environment variable, or connector secret. If authentication fails, including a tool result such as `Logged out`, tell the user to reconnect or update the Attribuly connector/API key in the host configuration. When the user asks how to configure Attribuly MCP, give these steps: 1. Create or locate an Attribuly API key from the user's Attribuly account or admin. 2. Store it outside the conversation, preferably as an environment variable named `ATTRIBULY_API_KEY`. 3. Add this MCP configuration to `~/.codex/config.toml`: ```toml [mcp_servers.attribuly] url = "https://data.api.attribuly.com/mcp" [mcp_servers.attribuly.env_http_headers] ApiKey = "ATTRIBULY_API_KEY" ``` 4. Restart Codex or reload MCP connectors so the new server and environment variable are available. 5. Test with a small query such as "Use $attribuly-mcp to show channel performance for the last 7 days." If the user's MCP host supports direct HTTP headers instead of environment-backed headers, the header name is `ApiKey`. Still recommend environment variables or a local secret manager rather than committing or pasting the real key. If no Attribuly MCP tools are available, stop before inventing data and tell the user that the connector must be enabled for this task. ## Workflow 1. Identify the business question, date range, attribution model, conversion goal, platform/channel, and granularity needed. 2. If the user omits the attribution model or conversion goal, default to model `linear` and goal `purchase`. Tell the user these defaults in the answer. 3. If the user omits dates for a performance request, use the last 30 complete days unless a daily, weekly, or month-to-date framing is clearly implied. State the exact date range in the answer. 4. Inspect the available Attribuly MCP tool names and schemas before calling them. Use returned schema fields rather than assuming parameter names. 5. Query the narrowest report that can answer the question. Paginate or refine filters when the tool returns partial or oversized results. 6. Validate that totals, filters, date range, attribution model, conversion goal, and currency are consistent before drawing conclusions. 7. Summarize in the user's language. Lead with the business takeaway, then include the supporting metrics, risks, and recommended next actions. ## Reporting playbook Read [references/reporting-playbook.md](references/reporting-playbook.md) when the request needs report routing, tool selection, metric interpretation, optimization advice, discrepancy diagnosis, or a structured output format. Read [references/marketing-action-playbook.md](references/marketing-action-playbook.md) when the user asks for recommendations, experiments, creative iteration, campaign optimization, CRO, influencer/referral analysis, offer/pricing strategy, lifecycle fixes, or a marketing plan from Attribuly data. Read [references/marketing-encyclopedia.md](references/marketing-encyclopedia.md) when the user asks for broad marketing strategy, channel planning, SEO/AEO/GEO, content operations, short-form/video/podcast repurposing, outbound or lead-generation strategy, revenue intelligence, sales/pricing support, marketing finance, or a "what else can we do?" answer that goes beyond the Attribuly report surface. ## Analysis rules - State the date range, attribution model, conversion goal, and report type used in every data-backed answer. - If the model or goal was not specified by the user, explicitly say that the current defaults are model `linear` and goal `purchase`. - Cite concrete returned metrics such as revenue, spend, orders/conversions, ROAS, CPA, MER, AOV, new-customer revenue, LTV, CTR, CVR, or attributed customers. - Avoid inventing metrics, benchmarks, platform spend, or customer context that is not returned by the MCP tools or supplied by the user. - Compare platform-reported performance with Attribuly performance before recommending major pauses, budget cuts, or scale-ups. - Treat budget changes, campaign pauses, tracking changes, and bid strategy changes as recommendations requiring human approval. - Flag empty data, missing integrations, attribution-window mismatches, incomplete spend, or partial-platform coverage as caveats. - For recurring reporting, keep a consistent table structure so week-over-week and month-over-month comparisons remain readable.