v1.0.0

First release: a Model Context Protocol server for Mathpix OCR over stdio and
Streamable HTTP, with every tool verified end-to-end against the live Mathpix
API.

Added

• Nine MCP tools: convert_document, convert_image, convert_strokes,
  batch_convert, check_document_status, search_results, get_usage,
  get_account_info, list_formats.
• Document conversion (PDF/DOCX/PPTX) via /v3/pdf for remote URLs and local
  files (multipart upload), returning Markdown/HTML.
• Image OCR via /v3/text, handwritten strokes via /v3/strokes, and recent
  captures via /v3/ocr-results; usage/account identifiers via /v3/ocr-usage.
• File-routed output: OCR results (LaTeX/text/MathML/HTML/bounding-box data) are
  always written to files and returned as a path + short preview + metadata, so
  results never overflow the model context. Destination follows the tool's
  output_path/output_dir, else MATHPIX_OUTPUT_DIR, else the system temp dir.
• Descriptive API errors that surface Mathpix's error_info.message (status and
  details included).
• Concurrent batch_convert (bounded thread pool) and polling that tolerates all
  intermediate Mathpix states.
• stdio transport (mathpix-mcp) and Streamable HTTP transport
  (mathpix-mcp-http / config.ru) with required bearer-token auth
  (MATHPIX_MCP_TOKEN).
• Configuration via environment variables (documented in the README and
  .env.example), including MATHPIX_OUTPUT_DIR for the output directory.
• Docs: README with an env-var reference and per-agent client setup,
  docs/DEPLOYMENT.md for HTTP deployment, MAINTAINING.md for release/admin,
  and a Dockerfile.
• Tooling/CI: RuboCop + RSpec (rake), GitHub Actions CI (rubocop, rspec,
  bundler-audit), Dependabot, and a trusted-publishing release workflow.

Dependencies

• Runtime: base64, mcp (>= 0.9.2), rack, puma (>= 8.0.2).