github johnkerl/miller v6.20.0

pre-release2 hours ago

What's new in 6.20.0

Miller and AI

Miller now treats AI agents as first-class users. Everything an agent needs to drive Miller well — discovering what exists, learning your data's shape, validating expressions before running them, and recovering from errors — is now a built-in, structured feature of the mlr binary, packaged for MCP-speaking agents (Claude Code, Claude Desktop, Cursor, ...) via a one-line setup:

claude mcp add miller -- mlr mcp

See the new Miller and AI docs page for the quick start. The pieces, all usable individually from the command line:

  • Machine-readable help catalog, mlr help --as-json: by @johnkerl in #2099, #2106
  • JSON capability index and mlr which intent router by @johnkerl in #2107
  • Structured verb options, with enum value-sets by @johnkerl in #2111
  • Structured error output, --errors-json, with hints and did-you-mean suggestions by @johnkerl in #2113
  • DSL validate/dry-run, mlr put --explain / mlr filter --explain by @johnkerl in #2131
  • New mlr describe verb: field names, types, cardinality, and value domains of your data in one pass by @johnkerl in #2132
  • New mlr mcp MCP server and agent playbook, with a --no-shell sandbox flag by @johnkerl in #2133
  • Roadmap doc by @johnkerl in #2103; "Miller and AI" docs page by @johnkerl in #2134

A first-class bytes type in the DSL

  • New bytes type with b"..." literals and base64/hex codecs by @johnkerl in #2122

More features

Bug fixes

  • Fix masked unset-on-array error path (along with govet lint findings) by @johnkerl in #2129

Documentation

Internals

New Contributors

Full Changelog: v6.19.0...v6.20.0

Don't miss a new miller release

NewReleases is sending notifications on new releases.