Research Ingestion: File a Source Into Your Knowledge Base

The Stack

Steps

1. 1

   Check your toolchain

   Config validation needs Node; Claude Desktop is the agent for the final step.

   command -v node >/dev/null 2>&1 || { echo "MISSING required: node"; exit 1; }
   command -v claude >/dev/null 2>&1 || echo "optional: Claude Desktop (for the ingestion step)"
   echo "preflight OK"

2. 2

   Validate the Claude Desktop config (Option A)

   Write the exact Option A block and prove it parses and points npx at obsidian-mcp.

   cat > claude_desktop_config.json <<'EOF'
   {
     "mcpServers": {
       "obsidian": {
         "command": "npx",
         "args": ["-y", "obsidian-mcp", "/Users/you/Documents/MyVault"]
       }
     }
   }
   EOF
   node -e 'const c=JSON.parse(require("fs").readFileSync("claude_desktop_config.json","utf8"));const o=c.mcpServers.obsidian;if(o.command==="npx"&&o.args.includes("obsidian-mcp")){console.log("config OK: npx -> obsidian-mcp")}else{console.log("BAD");process.exit(1)}'

3. 3

   Scaffold the inbox and references folders

   Create Inbox/ and References/ and drop a sample source note for Claude to file.

   mkdir -p vault/Inbox vault/References
   printf '%s\n' 'Source: a paper arguing vectorless RAG beats embeddings for small corpora.' > vault/Inbox/source.md
   echo "references ready: vault/References/"

4. 4

   File the source (the Claude step, not checked by CI)

   Tell Claude: read the note in Inbox/, create a summary in References/ with key insights + source metadata + 3 to 5 takeaways, then search the vault for related notes, link both ways, and flag anything that contradicts existing notes. Non-deterministic and key-gated, so CI doesn't run it.