clawmemory-mcp

An MCP server that lets claude.ai append durable memory entries to Aayush's Obsidian vault on GitHub.

What it does

Exposes three tools to claude.ai over the streamable-HTTP MCP protocol:

Tool	Purpose
list_vault_files	List .md files in a vault directory — used for dedup planning
read_vault_file	Fetch existing content of a vault file — used for fact-level dedup
append_to_vault	Append a structured markdown entry to a file (creates if missing). Server reads-then-writes safely with the existing sha — never overwrites blind

The GitHub PAT lives as an encrypted env var on Vercel; it never appears in any chat or claude.ai project setting.

Endpoints

• POST /api/mcp — the MCP endpoint claude.ai talks to. Requires Authorization: Bearer <MCP_BEARER_TOKEN>.

Local dev

npm install
cp .env.local.example .env.local   # fill in the values
npx vercel dev

Deploy

npx vercel link
npx vercel env add GITHUB_TOKEN production
npx vercel env add GITHUB_REPO_OWNER production
npx vercel env add GITHUB_REPO_NAME production
npx vercel env add GITHUB_BRANCH production
npx vercel env add MCP_BEARER_TOKEN production
npx vercel deploy --prod

Wire up claude.ai

1. Settings → Connectors → Add custom connector
2. Name: ClawMemory
3. URL: https://<your-vercel-deployment>.vercel.app/api/mcp
4. Auth: Bearer, value = MCP_BEARER_TOKEN
5. Add connector to the Brain project

After that, in any chat in the Brain project, claude.ai will see list_vault_files, read_vault_file, append_to_vault as available tools. Just type save to brain.