Claude can build anything.
Give it a place to save the data.

The part Claude can't do alone

Ask Claude Code for a habit tracker and you'll have working code in minutes — storing to a local file inside the session's sandbox. Close the terminal and the data is stranded; open the project on another machine and it's gone. The same goes for the artifacts Claude writes in a chat: perfect for showing, useless for keeping.

The missing piece is small and well-defined: a URL that accepts data and keeps it. FormLM's Data API is exactly that — a hosted, tokenized endpoint Claude can POST to from any page, script, or artifact it writes, and read back later through a query API. You paste one URL; Claude handles the rest, because the endpoint documents itself in Markdown and Claude reads documentation for a living.

Data API in action — publish, POST, and query back

Two ways to connect Claude

Hand Claude the endpoint URL. Create a form in FormLM, enable the Data API on the Publish tab, and paste the endpoint into the chat — with ?help appended. That URL returns a plain Markdown document listing every field, method, and example. Claude reads it, asks nothing, and starts POSTing valid records. No MCP, no CLI, no config file:

"Store your outputs here instead of writing files:
 https://formlm.me/api/v3/share/YOUR_TOKEN?help
 Read the documentation first, then submit each result."

# Claude fetches the Markdown docs, matches the field names,
# and POSTs structured records — on its own, every time.

Connect formlm-cli as an MCP server. Now Claude doesn't just store data — it can create forms, define fields, publish endpoints, and inspect results conversationally, with FormLM's domain rules embedded in every tool description. Setup differs slightly between the two Claude apps:

Claude Code — one command in your project:

$ claude mcp add formlm -- formlm-cli mcp
# starts the FormLM MCP server (8 tools + 6 resources)
# auth happens in-chat on first use: paste your Access Token (Account Settings), or email verification code, or email + password

Claude Desktop — edit the config file, then fully restart the app (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json, Windows: %APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "formlm": {
      "command": "formlm-cli",
      "args": ["mcp"],
      "env": {
        "PATH": "/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin"
      }
    }
  }
}

The env.PATH line matters on macOS: Claude Desktop launches MCP servers without your terminal's PATH, so formlm-cli — installed via Homebrew Node — would otherwise fail with command not found. If it still fails, run which formlm-cli and use the absolute path as the command. Install the CLI first with npm install -g @formlm/cli; the full walkthrough lives in the developer zone.

What it looks like in practice

$ curl -X POST https://formlm.me/api/v3/share/YOUR_TOKEN \
  -H "Content-Type: application/json" \
  -d '{
    "data": {
      "topic": "vector-db benchmarks",
      "verdict": "pgvector sufficient under 1M rows",
      "confidence": "high"
    },
    "meta": {"source": "claude-research-session"}
  }'

# Read — next session, Claude recalls everything it found
$ curl https://formlm.me/api/v3/share/YOUR_TOKEN/query?page=1&size=50

That's the whole pattern: persistent memory with zero infrastructure. CORS is enabled on reads and writes, so an artifact, a local HTML page, or a script Claude writes in the sandbox can all reach the endpoint. Stored records are also visible in FormLM's UI and exportable as CSV — the data belongs to you, not to the chat.

Frequently asked questions

Use it with other agents