v2.1.0

v2.1.0

81 changes since the previous release.

New Features

• re-add plotly package as a runtime dependency (maintainer request)
• add pixi engine-free job, gate docker build behind license flag, re-home publish smoke onto pip
• add CI version-drift guard and clean-venv offline smoke test
• rebuild vendor/ wheelhouse to the fastmcp 3.4.4 closure
• fix update_vendor.bat and regenerate requirements-lock.txt to fastmcp 3.4.4 closure
• rewrite docker-compose to network-license mathworks/matlab story
• rewrite Dockerfile to honest mathworks/matlab + network license
• emit structured output + ToolResult(is_error=True) on MATLAB errors
• add ExecuteResult/CheckResult schema-additive models
• generate and commit pixi.lock
• add pixi.toml workspace manifest with matlabengine deliberately absent
• delegate MATLAB engine install to shared script
• register TimingMiddleware in create_server behind monitoring flag
• add TimingMiddleware for per-tool-call metrics
• bump fastmcp to 3.4.4 and drop unused plotly runtime dep
• implement scripts/install_matlab_engine.py — GREEN
• add scale-down tests, replace sleep races, strengthen assertions
• convert asyncio.run tests to async def, consolidate mock_pool fixture
• fix server.py deprecations, session errors, CORS wiring, double-timeout docs
• fix config model — dead attrs, YAML errors, CORS origins, inspect_mode
• background task tracking, executor workspace API, shutdown
• add job state transition guards and TOCTOU-safe session creation
• add workspace API and _needs_replacement flag to MatlabEngineWrapper
• wire security validator into JobExecutor in server.py
• add security parameter to JobExecutor and centralize check_code

Bug Fixes

• lean sdist + changelog that never hard-fails
• clear pre-existing ruff F401/F841 in test files
• add pip to pixi env so pixi run install-engine works
• stop generic CI building the licensed image for the trivial client (#2)
• map MATLAB_RELEASE to matching matlabengine pin in Dockerfile (#3)
• export detected MATLAB_RELEASE to the engine script (#4)
• let offline install bypass whitelist; catch pip failures in main (#5, #6)
• match 6-char MATLAB release dirs in install glob (#1)
• drop blanket --trusted-host, echo resolved dirs, reconcile banner
• close verification gaps — replace last asyncio.sleep, mark plans complete
• update _make_mock_pool references to conftest make_mock_pool after merge
• fix engine release leak, start timeout, acquire race, and status count
• session ID sanitization against path traversal
• expand blocklist, fix empty-token auth, add security-disable warning
• deduplicate route handlers, fix error count, prime cpu_percent
• add SQL LIMIT, clamp query params, fix path traversal

Documentation

• rework all wiki pages from the current codebase
• record phase executed, verification deferred to user's Win10/MATLAB env
• phase context, research, validation, verification (human_needed) + plan progress
• add plan summary for CI drift guard + offline smoke + pixi + license-gated docker
• add plan summary for conda deprecation + wiki reconciliation
• reconcile wiki Installation.md to R2022b+ and pixi-first
• deprecate conda env files toward pixi
• complete vendor rebuild + update_vendor.bat fix plan summary
• complete Docker rewrite to honest mathworks/matlab plan
• complete structured output for execute_code/check_code plan
• complete pixi manifest/lock/README plan summary
• add pixi quickstart and streamablehttp/sse transport guidance
• complete install.bat delegation plan summary
• complete TimingMiddleware for per-tool-call metrics plan
• complete FastMCP 3.4.4 bump + plotly cleanup plan
• reconcile FastMCP version string in CLAUDE.md to 3.4.4
• complete MATLAB engine install script plan
• create phase plan — 10 plans in 3 waves (pixi/docker/fastmcp 3.4.4)
• mark phase complete — 7/7 plans, 39 issues resolved
• complete test quality improvements plan summary and state update
• add self-check result to SUMMARY
• complete server and config cleanup plan summary and state update
• complete job/session state machine fixes plan
• complete pool resource safety plan summary and state update
• complete centralize-security-in-executor plan
• complete security hardening plan summary and state update
• complete monitoring subsystem security and accuracy fixes plan
• create phase plan — 7 plans across 4 waves for 39 codebase review issues
• add validation strategy and update roadmap for phase 7
• research phase 7 fix all HIGH and MEDIUM codebase review issues
• auto-generated context (infrastructure phase)

CI/CD

• remove AI wiki-gen workflow (docs.yml)
• exercise the real MATLAB release-detection path (no license needed)

Other Changes

• add failing tests for MATLAB engine install script
• Merge pull request #7 from HanSur94/claude/fix-release-pipeline
• Merge pull request #6 from HanSur94/claude/release-2.1.0
• bump version to 2.1.0
• Merge pull request #5 from HanSur94/claude/rework-wiki
• Merge pull request #4 from HanSur94/claude/win10-install-lean-3d6b8f
• Create SECURITY.md

v2.0.2 — Conda/Miniconda Support

What's Changed

Added Anaconda/Miniconda installation support — users can now set up the server with a single conda command.

Conda Support

For end users (no repo clone needed):

conda env create -f environment.yml
conda activate matlab-mcp
matlab-mcp --inspect

For developers (from repo clone):

conda env create -f environment-dev.yml
conda activate matlab-mcp-dev
pytest tests/ -x -q

CI

• conda-test-pypi — validates environment.yml installs from PyPI and server loads
• conda-test-dev — validates environment-dev.yml installs from source and all tests pass
• 19 CI jobs total across Linux, Windows, macOS, Docker, conda

Full Changelog

v2.0.1...v2.0.2

v2.0.1 — CI & Integration Testing

What's Changed

Post-milestone improvements to CI pipeline and test coverage. No production code changes except one bug fix (monitoring routes on streamablehttp).

Bug Fixes

• Monitoring routes on streamablehttp — /dashboard, /health, /metrics now registered for the new transport (was SSE-only)
• 12 ruff lint errors — unused imports and f-string placeholders cleaned up

CI & Testing

• MCP integration test — live server subprocess + MCP client on localhost (Linux + Windows)
• Docker Compose remote test — client container connects to server container over Docker network with bearer auth
• Windows install.bat smoke test — validates the real user install path produces a working installation
• 17 CI jobs across Linux, Windows (2022 + latest), macOS, Docker

Full Changelog

v2.0...v2.0.1

v1.5.0

What's New

Inspect Mode

• Added --inspect flag to start the server without MATLAB for tool discovery and CI inspection

Glama Integration

• Added glama.json for MCP server listing metadata
• Server is now listed and inspectable on Glama

Security

• Removed unused diskcache dependency (medium severity vulnerability)

v1.4.1

v1.4.1 — Documentation

Comprehensive Docstrings

Improved documentation across all source modules:

• server.py: MatlabMCPServer.__init__, lifespan context manager, main entry point, background tasks
• pool/engine.py: EngineState lifecycle, state property
• pool/manager.py: _make_engine, get_status return dict
• jobs/models.py: JobStatus terminal states, Job attributes
• jobs/executor.py: _inject_job_context, _wait_for_completion, _safe_serialize
• jobs/tracker.py: thread safety, pruning strategy
• monitoring/: collector init, store init, health evaluation, route handlers, dashboard factory + all inner handlers
• output/plotly_style_mapper.py: mapping tables, all trace converters
• tools/monitoring.py: all three tool functions with Parameters/Returns
• tools/custom.py: inner handler docstring
• config.py: all 11 Pydantic models, load_config, validate_pool

README Updates

• Added Windows one-click installer section (install.bat)
• Updated MATLAB version requirement to R2022b+
• Added offline install instructions

Full Changelog: v1.4.0...v1.4.1

v1.4.0

What's New in v1.4.0

One-Click Windows Installer

• install.bat — fully offline, no-admin-rights installer for Windows 10/11
• Bundles all 100+ Python wheels in vendor/ for zero-network installation
• Auto-detects MATLAB R2022b+ and Python 3.10/3.11/3.12
• Handles corporate SSL proxies, ancient pip versions, and read-only Program Files
• Generates start-matlab-mcp.bat launcher and prints ready-to-paste config for Claude Desktop, Claude Code, and Cursor

MATLAB Engine API Install Improvements

• Copies Engine source to temp with directory junctions to avoid access-denied errors in Program Files
• Patches _arch.txt to point to real MATLAB after temp-dir build
• Works fully offline with --no-build-isolation

CI/CD

• Added Windows test matrix: windows-2022 (Win10) + windows-latest (Win11) × Python 3.10 + 3.12
• Cross-platform test fixes for Windows path handling

Bug Fixes

• Fixed .gitignore excluding monitoring/ subpackage from wheel
• Fixed unused imports in monitoring/dashboard.py
• Added missing Windows-only dependencies: pywin32, pywin32-ctypes, colorama, tzdata, async-timeout, tomli, taskgroup

Full Changelog: v1.3.0...v1.4.0

v1.3.0

What's Changed

Test Coverage

• Increased code coverage from 63% to 90% with 385 new tests across 11 test files
• Added comprehensive tests for all previously untested modules:
  • tools/jobs, tools/discovery, tools/custom, tools/files, tools/admin
  • server (MatlabMCPServer, create_server, main)
  • monitoring/dashboard, monitoring/store, monitoring/collector
  • jobs/executor, output/formatter, output/thumbnail
  • session/manager, pool/engine, monitoring/health
• 755 tests now passing (up from 378)

Modules at 100% Coverage

tools/jobs, tools/discovery, tools/admin, tools/core, jobs/models, jobs/tracker, monitoring/health, monitoring/routes, output/formatter, output/thumbnail, session/manager

v1.2.0

v1.2.0

This release introduces automated wiki generation and enhances documentation with MATLAB-to-Plotly plot conversion features.

New Features

• Auto-generate all 11 wiki pages via Claude API on push to master

Documentation

• Add MATLAB-to-Plotly plot conversion feature section to README
• Add Plotly conversion screenshot to README
• Add comprehensive specification for full wiki auto-generation with file mappings, validation, and error handling

v1.0.1 — CI fixes & security hardening

What's Changed

Security

• Fixed incomplete blocklist in shipped config.yaml — all 11 blocked functions (system, unix, dos, !, eval, feval, evalc, evalin, assignin, perl, python) are now listed, matching code defaults and documentation

Fixes

• Fixed 35 ruff lint errors (unused imports/variables) across src/ and tests/
• Bumped minimum Python from 3.9 to 3.10 (fastmcp requires 3.10+)
• Added psutil and uvicorn to [dev] extras so monitoring works out of the box
• Fixed get_error_log docs to show actual default (20, not "varies")
• Fixed README workspace isolation description to include all 5 cleanup commands
• Added PyPI keywords and classifiers for better discoverability

Upgrade

pip install --upgrade matlab-mcp-python

Full changelog: v1.0.0...v1.0.1

v1.0.0 — First stable release

MATLAB MCP Server v1.0.0

Connect any AI agent (Claude, Cursor, Copilot) to MATLAB via the Model Context Protocol.

Highlights

• 20 MCP tools — execute code, async jobs, file I/O, toolbox discovery, code checking, monitoring
• Elastic engine pool — scales 2–10+ MATLAB engines based on demand
• Interactive Plotly plots — MATLAB figures auto-converted to Plotly JSON with WebGL for large datasets
• Async job system — long-running simulations with progress reporting
• Custom tools — expose your .m functions as first-class AI tools via YAML
• Monitoring dashboard — live gauges, time-series charts, execution log, health/metrics endpoints
• Security — function blocklist, workspace isolation, filename sanitization, upload limits
• Multi-user support — SSE transport with session isolation
• Docker support — Dockerfile and docker-compose.yml included
• Cross-platform — macOS, Windows, Linux (MATLAB 2020b+)

Install

pip install matlab-mcp-python

Requires MATLAB 2020b+ with the Engine API for Python installed.

Quick Start

# Run the server
matlab-mcp

# Connect to Claude Code
claude mcp add matlab -- matlab-mcp

See the README and Wiki for full documentation.