Skip to content

CHANGELOG.md

Latest commit

 

History

History
163 lines (132 loc) · 13.9 KB

CHANGELOG.md

File metadata and controls

163 lines (132 loc) · 13.9 KB
Raw

Changelog

All notable changes to MCP-Maker will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.7] - 2026-07-19

A full audit of all 13 connectors and their code-generation templates: 20+ bug fixes (several of them crashes in generated servers) and new query tools across every connector.

Added

  • mcp-maker chat now supports Grok (xAI) and DeepSeek — auto-detected from xai- key prefixes or the XAI_API_KEY / DEEPSEEK_API_KEY env vars, with --provider grok|deepseek for explicit selection.
  • distinct_{table} tool for SQLite, PostgreSQL, MySQL, MongoDB, Files, Excel, Google Sheets, Airtable, Notion, and Supabase — distinct values of any column/field/property.
  • aggregate_{table} tool for Files, Excel, Google Sheets, and Airtable (count/sum/avg/min/max with GROUP BY semantics) — SQL and MongoDB already had it.
  • filter_{table} tool for Files, Excel, and Google Sheets — operator-based filtering (eq/ne/gt/gte/lt/lte/contains).
  • View discovery for SQLite, PostgreSQL (incl. materialized views), and MySQL — views get read tools and are marked read-only.
  • Single-file mode — mcp-maker init ./users.csv works directly on a CSV/TSV/JSON/JSONL file, no directory needed.
  • TSV and JSONL support in the Files connector (JSONL previously failed silently and became a blob resource).
  • MongoDB Atlas — mongodb+srv:// connection strings are now registered.
  • OpenAPI: $ref schema resolution, Swagger 2.x in: body parameters, path-level shared parameters, a generic call_api escape-hatch tool, and three auth styles (API_TOKEN Bearer / API_KEY+API_KEY_HEADER / API_KEY+API_KEY_QUERY).
  • Redis: key_info (type/TTL/memory), scan_keys (glob pattern scan), and set_expiry tools.
  • Google Sheets: numeric-aware sort_field/sort_direction on list_.
  • Template render test suite — every connector's templates are rendered with synthetic schemas and syntax-checked (39 new test cases), plus new connector tests (367 total).

Fixed

  • Notion generation crashed entirely — an unbalanced Jinja block in the Notion template raised TemplateSyntaxError on every mcp-maker init notion://….
  • Files servers: count_, search_, and get_by_row were broken — they treated the paginated dict returned by list_ as a list (count always returned 4; search/get crashed or returned garbage).
  • Airtable search_ crashed at runtime for the same dict-vs-list reason.
  • Google Sheets count_ returned the grid size (often the default 1000) instead of actual data rows; sheets with duplicate/empty headers broke reads.
  • Supabase search_ required every text column to match (chained .or_() calls are ANDed) — now matches any column; validate() no longer reports success with a wrong API key.
  • PostgreSQL ?schema= URI option broke the connection (passed through to libpq as an unknown option).
  • MySQL passwords with special characters failed (credentials weren't URL-decoded); empty database names now raise a clear error; connections use utf8mb4.
  • Redis dropped the username from ACL-style URIs; connection params (including password) are no longer stored in schema metadata.
  • Excel dates/datetimes were typed as strings; duplicate headers collided; generated data keys didn't match tool schemas after sanitization; int+float columns are now FLOAT.
  • Airtable/Notion init no longer downloads entire bases just to count rows (capped at 1,000 records / 10 pages).
  • HubSpot custom objects with special characters produced invalid tools calling wrong endpoints; failed counts report unknown instead of 0.
  • MongoDB collection names are sanitized (with a name map so queries still hit the real collection); a null in the first sampled document no longer locks a field's type to unknown.
  • SQLite inspection opens the database read-only and safely escapes quoted identifiers.
  • Registry: gsheet://, hubspot://, and mongodb+srv:// URIs now give correct pip install hints when dependencies are missing.

[0.2.6] - 2026-03-13

Added

  • mcp-maker chat command — Interactive terminal chat with your SQLite database via natural language. Uses LLM function calling to auto-generate tools from your schema and execute queries.
  • OpenAI & OpenRouter support — Chat supports both OpenAI and OpenRouter (500+ models including Claude, Gemini, Llama, DeepSeek). Provider auto-detected from API key prefix (sk-or- → OpenRouter).
  • core/agent.py — New module: schema_to_tools() converter, QueryExecutor for SQL execution, ChatAgent for multi-turn LLM conversations with tool calling.
  • cli/chat.py — Rich-formatted REPL with tool call transparency, built-in commands (schema, clear, exit).
  • chat optional dependency — pip install "mcp-maker[chat]" installs openai>=1.0.0.
  • 35 new tests — Comprehensive coverage for schema-to-tools conversion, query execution, and ChatAgent with mocked LLM (307 total tests).

Fixed

  • LLM API error messages — Errors from OpenAI/OpenRouter now show the actual error message (e.g., "Invalid API key", "model not found") instead of raw HTTP status codes.

Changed

  • README rewritten — Professional open-source documentation with two-path Quick Start, generated tools table, security section, MCP client compatibility table.

[0.2.5] - 2026-03-13

Fixed

  • Critical: Insert function parameter ordering — Schemas with nullable columns before non-nullable columns (e.g., foreign keys before NOT NULL fields) caused SyntaxError: parameter without a default follows parameter with a default in generated server code. Fixed across all 6 SQL templates (SQLite, PostgreSQL, MySQL + async variants) by sorting required parameters before optional ones.
  • Codecov CI integration — Added CODECOV_TOKEN secret and upgraded to codecov-action@v5.

[0.2.4] - 2026-03-13

Security

  • SQL Injection Prevention: Added column whitelist validation (_KNOWN_COLUMNS + _validate_column) in consolidated-mode tools across SQLite, PostgreSQL, and MySQL templates.
  • Credential Hardening: Removed hardcoded DB_PATH fallback in SQLite template; now raises RuntimeError if DATABASE_URL is unset.
  • API Key Authentication: Auth wrapper now validates a client-provided api_key parameter against MCP_API_KEY environment variable, replacing the previous no-op check.
  • Token Leak Prevention: HubSpot connector now sanitizes URIs in error messages to prevent API token exposure.
  • Batch Input Sanitization: All batch_insert_ tools now validate column names against schema whitelist before SQL interpolation.

Fixed

  • Connection Leaks: PostgreSQL cursor cleanup moved to finally blocks. SQLite connections now registered with atexit handler for shutdown cleanup.
  • Batch Size Limits: Batch insert/delete operations capped at 1,000 records to prevent resource exhaustion.
  • CLI Flag Conflict: Warning now emitted when both --read-write and --ops flags are provided.
  • Mixed Source Types: Multi-source init now supports mixing source types (e.g., SQLite + MongoDB) with a warning instead of rejecting.
  • .env Parser: Now correctly handles export prefix and strips inline # comments from unquoted values.
  • Connector Registration: Added missing optional connectors (Excel, MongoDB, Supabase, OpenAPI, Redis) to connectors/__init__.py.
  • Connector Loading: Improved error handling to distinguish missing third-party dependencies from internal import errors.

Changed

  • Project Status: Upgraded from Alpha to Beta (Development Status :: 4 - Beta).
  • Architecture: Extracted cli/connectors_loader.py and cli/schema_ops.py from generator.py, reducing it from 626 to 512 lines.
  • Connector Naming: Renamed redis_connector.py to redis.py for consistency with all other connectors.
  • Test Organization: Moved test_cloud_inspect.py and test_sql_inspect.py to tests/core/. Renamed test_env_extended.py to test_env_edge_cases.py. Added conftest.py and __init__.py to all test subdirectories.
  • Documentation: Deleted root-level DOCS.md in favor of docs/reference.md. Expanded docs/hubspot.md from 95 to 250+ lines.
  • CI/CD: Added Codecov upload, coverage badge, ruff and mypy configuration in pyproject.toml.
  • Repository Hygiene: Removed tracked generated files and test artifacts from git.

Added

  • SECURITY.md: Vulnerability reporting policy with 48-hour response SLA.
  • CODE_OF_CONDUCT.md: Contributor Covenant v2.1.
  • Makefile: Standard make test, make lint, make fmt, make clean targets.
  • Test Coverage: 272 tests passing (up from 256), including 22 new regression tests for security and correctness fixes.

[0.2.3] - 2026-02-23

Added

  • HubSpot Data-Bridge: Natively connects to HubSpot CRM using Private App Tokens (hubspot://pat=YOUR_TOKEN).
  • Deep Auto-Discovery: Automatically traverses and introspects all Custom Properties for Contacts, Companies, Deals, Tickets, Products, Quotes, Notes, Tasks, Meetings, Emails, Calls, and any defined Custom Objects natively.
  • Context Maximization: Advanced Jinja generation intelligently maps human names (hubspot_get_owners), sales stages (hubspot_get_deal_pipelines), and audience segments (hubspot_get_lists).
  • Compound Tools: Advanced tools like hubspot_search_crm_objects and batch_upsert_{table} created for efficient data-syncing natively.

[0.2.2] - 2026-02-23

Added

  • Pagination & Sorting: All list_ tools now return {results, total, has_more, next_offset} with sort_field and sort_direction params across every connector (SQL, NoSQL, Excel, Files, GSheets, Notion, Airtable, Supabase).
  • Field Selection: list_ tools accept fields="col1,col2" for column projection with SQL injection-safe whitelisting.
  • Date Range Filters: Auto-detected date/datetime columns get date_from and date_to params on list_ tools.
  • Foreign Key Discovery & Join Tools: Auto-discovers FK relationships in SQLite, PostgreSQL, and MySQL. Generates join_{from}_{to}() tools for cross-table queries.
  • Batch Operations: batch_insert_{table}(records=[...]) and batch_delete_{table}(ids=[...]) tools with transaction safety.
  • Export Tools: export_{table}_csv() and export_{table}_json() for every table across all connectors.
  • Webhook Support: Optional --webhooks flag generates register_webhook_{table}(), list_webhooks_{table}(), and fire_webhooks_{table}() tools.
  • Redis Pub/Sub: publish_message(channel, message), channel_list(pattern), and channel_subscribers(channel) tools.
  • Multi-Source Initialization: mcp-maker init source1 source2 ... merges schemas from multiple data sources into a single unified MCP server.
  • Async Templates: --async flag generates aiosqlite-based async handlers for SQLite connectors.

Changed

  • Connector Documentation: All 11 connector docs updated with Advanced Features sections covering pagination, fields, date filters, batch, export, and join tools.
  • Test Coverage: Boosted from 43% to 73% (244 tests) with comprehensive mock-based tests for Excel, MongoDB, Redis, Supabase, OpenAPI, MySQL, PostgreSQL, GSheets, Notion, Airtable connectors and CLI deploy/server/environment commands.
  • Lint Clean: Resolved all 402 ruff lint errors (F401, F811, F841, E402).

[0.2.1] - 2026-02-23

Added

  • Granular Security (--ops): Control exactly what the LLM can edit with precise --ops read,insert,update,delete arrays. 100% template-native conditional compilation.
  • Cloud API Rate Limiting: Extensively upgraded the server.py.jinja2 generation engine to auto-inject a TokenBucketRateLimiter securing Cloud endpoints (Notion, Airtable, Sheets) against 429 Too Many Requests bans.
  • Google Sheets Connector: Direct integration with gsheets://URL endpoints. Generates update_cell tools.
  • Notion Connector: Supports Multi-Database URIs natively, handles cursor pagination, auto-parses 20+ Notion property types.
  • E2E Test Framework: Deployed a dynamic End-to-End server generation and mock execution testing pipeline in tests/e2e.

Changed

  • Massive Architectural Refactor: Obliterated monolithic dependencies. cli.py and test_mcpforge.py bisected gracefully into modular domain packages.
  • Test Integrity: Test suite significantly expanded (88 passing regression checks).
  • Re-architected Open Source Documentation (CONTRIBUTING.md) for streamlined Fork/Branch/PR Git workflows.

[0.2.0] - 2026-02-21

Added

  • PostgreSQL Connector: Fully typed extraction for postgres schema inspection.
  • MySQL Connector: Queue-based connection pools supporting mysql://.
  • Airtable Connector: Auto-generation of formula-based queries and Base Discovery via mcp-maker bases.
  • Semantic Search: Vector database integrations explicitly enabled via --semantic.

0.1.0 - 2026-02-21

Added

  • CLI with 4 commands: init, inspect, serve, list-connectors
  • SQLite connector: inspects tables, columns, types, primary keys, and row counts
  • File connector: CSV/JSON files → tables with type inference, text files → read-only resources
  • Code generator: produces complete FastMCP server with 5 tools per table
    • list_{table}(limit, offset) — paginated listing
    • get_{table}_by_{pk}(id) — lookup by primary key
    • search_{table}(query, limit) — text search across string columns
    • count_{table}() — total row count
    • schema_{table}() — column names and types
  • Connector registry pattern for easy community contributions
  • Rich terminal output with tables, panels, and status spinners
  • 24 unit tests covering schema inspection, connectors, and code generation
  • GitHub Actions CI/CD: tests on Python 3.10–3.12, auto-publish to PyPI on release