Agent Ready

AI agents use Finetype through Arcform — `arc mcp` serves Finetype's commands over the Model Context Protocol.

AI agents are good at running commands. They are less good at knowing which commands to run, in what order, and what to do when something unexpected happens.

An agent reaches Finetype through Arcform. Finetype v0.6.63 removed its own MCP server and its agent skills. Arcform's arc mcp now serves Finetype's commands to an agent over the Model Context Protocol, and Arcform's next release carries the skills (see Agent skills). An agent working through arc mcp leaves behind a Protocol, a folder holding arcform.yaml and its SQL, that anyone can run again with arc run.

Connect an agent

Install Finetype and Arcform, with the duckdb command on your PATH (the Arcform install notes say which DuckDB build it needs). Arcform installs with:

curl -fsSL https://install.meridian.online/arc | bash

Then add arc mcp to Claude Code:

claude mcp add arc -- arc mcp

Any other MCP client launches the same command as a stdio subprocess:

{
  "mcpServers": {
    "arc": {
      "command": "arc",
      "args": ["mcp"]
    }
  }
}

An agent started from a desktop app may not see your shell's PATH. In that case give the absolute path of arc (which arc prints it) as the command, and an env entry whose PATH holds duckdb, finetype and uv (uv runs Arcform's Python operators). Arcform's README has the Codex configuration too.

Finetype's tools

arc mcp serves five Finetype tools. Each runs the finetype binary on your machine, which must be Finetype v0.6.54 or later, so an agent and an analyst at the terminal see the same Finetype.

Tool	What it does
`infer`	Infer the semantic type of string values, through `finetype infer`
`profile`	Profile each column of a CSV or JSON file, through `finetype profile`. Pass `format: "json-schema"` for a table-level JSON Schema
`taxonomy`	Browse Finetype's type taxonomy, or export a type's JSON Schema
`validate`	Validate CSV or Parquet data against a JSON Schema and report its quality
`generate`	Generate synthetic sample data from Finetype's taxonomy

Beside them, arc mcp serves Arcform's own tools, so the same agent can create a Protocol, record SQL steps into it, run it and query the tables it makes.

If your client still runs `finetype mcp`

Since Finetype v0.6.63, finetype mcp exits with an error whose message names arc mcp, so a client still configured for it reports where to connect. Replace that entry with the arc one above.

Agent skills

Arcform's next release carries six agent skills that teach an agent the analyst's workflow through these tools, from agreeing the question to handing over the answer, along with an arc skills install command and a plugin for Claude Code, Codex, Gemini CLI and Cursor that installs them. Arcform v0.2.2, the latest release on 2026-10-27, does not have them yet. Until it does, an agent connected to arc mcp has the tools without the skills.

The pipeline in action

Here is the pipeline at the terminal, as finetype runs it:

# Step 1: Understand the data
finetype profile -f sales.csv

# Step 2: Capture the contract
finetype profile -f sales.csv -o json-schema > sales.schema.json

# Step 3: Validate & materialise — gate the data and cast it in one pass
finetype validate sales.csv sales.schema.json --db analytics.db --table sales