How a Marketing Consultant Uses Obsidian and Claude Code to Manage Dozens of Clients

A marketing consultant generating $40K per month in recurring revenue walks through his entire system for managing multiple clients using Obsidian as a visual layer and Claude Code as the execution engine. The core insight: AI tools are stateless, and Obsidian becomes the persistent memory that makes Claude Code truly useful across sessions.

The Core Problem: Statelessness

AI tools do not remember anything between sessions. Even with desktop app memory or project-level context, you cannot get as intentional as you can with markdown files structured in an Obsidian vault. The vault becomes Claude's memory -- a persistent, structured, version-controlled knowledge base that the AI reads directly from your filesystem without copy-pasting or document uploads.

System Architecture

The system has two layers connected by Obsidian's wiki links:

Layer	Tool	Role
Visual	Obsidian	Renders dashboards, pointer notes, weekly plans, review files
Execution	Claude Code	Reads and writes local files directly from the terminal
Connective	Wiki Links	Bidirectional links between notes for navigation

Claude Code lives in the terminal, reads the vault files, does the work, and updates them when finished. Wiki links let both the human and the AI navigate between clients, tasks, and deliverables without ever searching for a file.

Vault Structure: The Three Non-Negotiables

Every client gets a folder with a consistent internal structure. That consistency is what makes the system scale.

1. Context File

Written once during onboarding. Contains brand voice, services offered, target market, SEO requirements -- everything that defines who the client is. Every time Claude runs a skill or references that client, it reads this file first. You never re-explain context.

2. Log File

A chronological, timestamped record of deliverables and client interactions: SEO implementations, metadata optimizations, Google Ads refreshes, even client emails. Skills write entries automatically with date, action type, what happened, and what is next. It is a human-readable history of the entire client relationship.

3. YAML Trackers

Machine-readable files tracking every URL on a client's site with status (planned, in progress, done) plus a history of which skills touched it and when. Skills read the tracker to know what to work on, update it when work is done, and dashboards regenerate from the data. The YAML frontmatter at the top of every markdown file provides structured metadata like status, dates, categories, and client name.

The Weekly Plan Workflow

Each week, a skill called Weekly Plan triggers a multi-step process:

1. Claude reads the global index of all clients
2. Checks each client's tier and cadence (scope of agreement)
3. Calculates what work is due, what is overdue
4. Generates a prioritized work plan with two outputs: a human-readable checklist in Obsidian and a machine-readable run queue for automatic skill execution

Every client name in the plan is a wiki link to their pointer note. Every review file is linked. Navigation through the week's work requires zero searching.

The Wrap-Up Skill

After completing a typical four-skill SEO engagement flow, one command triggers the wrap-up skill that updates five surfaces:

1. Marks pages as done in the tracker
2. Logs the completed work
3. Regenerates the overview dashboard across all clients
4. Updates the client's pointer note
5. Drafts a client email in Gmail highlighting completed work

No manual bookkeeping. Nothing falls through the cracks because the system, not the consultant, keeps track of everything.

Scaling and Onboarding

Adding a new client takes minutes: create the folder, write the context file, run a setup tracking skill. All YAML frontmatter and tracking infrastructure is created by Claude Code at the onboarding stage. This is how one person manages dozens of clients without a large team.

The system also uses Firecrawl's API to scrape client domains (similar to Screaming Frog), generating a list of all URLs that gets matched against the SEO roadmap and scope of work.

Key Takeaways

• Obsidian is the persistent memory layer; Claude Code is the execution engine that reads and writes to it
• Consistent folder structure per client is what makes the system scalable
• YAML frontmatter provides machine-readable metadata that skills can parse programmatically
• One wrap-up command updates five different surfaces, eliminating manual bookkeeping
• Use Claude Code's plan mode when designing your vault structure -- be intentional about what you need
• Wiki links enable both human and AI navigation without file searching

Resources

• Obsidian -- Markdown-based knowledge management with wiki links and bidirectional connections
• Claude Code -- Anthropic's terminal-based AI coding agent
• Firecrawl -- Web scraping API for extracting site structure and URLs