Skip to content

Changelog

All notable changes to this project will be documented in this file.

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

[0.1.3] - 2026-03-09

Added

  • Schema cache TTL (CACHE_TTL setting) — Optional time-to-live (in seconds) for the schema cache; when set, the cached schema processor is automatically rebuilt after the TTL expires (default: None — no expiration, preserving existing behavior)
  • invalidate_schema_cache() function — Public, thread-safe API to force-clear the cached schema processor on demand (e.g., from a Django signal handler or management command); importable from drf_mcp_docs or drf_mcp_docs.server
  • Pagination-aware code generation — Detects DRF pagination patterns (page number, limit/offset, cursor) from OpenAPI response schemas and generates helper iterators that fetch all pages automatically for every supported client (fetch, axios, ky, requests, httpx)
  • cURL code generation — New curl output language for generate_code_snippet with proper method, auth headers, query params, and JSON body formatting; also accepts shell, sh, or bash as aliases
  • PaginationInfo dataclass — New frozen dataclass capturing pagination style, results field name, and count availability
  • checkmcpconfig management command — Validates settings (unknown keys with typo hints, transport, language, client compatibility, CACHE_TTL, path prefixes), tests adapter detection, and verifies schema generation before running the server
  • --reload flag on runmcpserver — Uses Django's built-in autoreload mechanism to restart the MCP server when source files change during development; only supported with streamable-http transport
  • Structured debug logging — Added logging.getLogger(__name__) to 11 modules covering adapter selection, schema processing, cache lifecycle, tool invocations, resource access, ASGI routing, and settings resolution; enable via Django's standard LOGGING config under the drf_mcp_docs namespace
  • Daphne ASGI server compatibility — Refactored mount_mcp() to directly manage the MCP session manager lifecycle instead of relying on Starlette's lifespan middleware, fixing the "Task group is not initialized" error on servers that do not send lifespan events
  • Detailed default server instructions — Replaced generic default instructions with structured workflow guidance describing all available resources and tools, helping AI agents effectively navigate the API documentation
  • MCP client integration tests — End-to-end tests that boot a real uvicorn server and connect an actual MCP client via streamable-http transport, catching protocol and transport-level issues that unit tests cannot detect
  • Cache invalidation tests — TTL expiration, manual invalidation, backward compatibility, top-level import, and concurrent invalidation stress test
  • Config validation tests — Comprehensive tests for checkmcpconfig covering unknown keys, typo detection, transport/language/client compatibility, and adapter verification
  • Logging tests — Tests verifying debug log output across all 11 instrumented modules

Fixed

  • list_schemas token overflow — Slimmed list_schemas output to return only schema name and description, preventing MCP token limit issues on large APIs; full schema details remain available via get_schema_detail

Documentation

  • Troubleshooting guide — Comprehensive guide covering adapter detection, schema generation, ASGI integration, connection/transport, configuration, caching, and debug logging, all in symptom-cause-solution format
  • Example bookstore project — Minimal Author/Book project with drf-spectacular and drf-mcp-docs pre-configured, including ASGI mount, both transport modes, and AI tool configuration snippets
  • Lifespan forwarding guide — Instructions for custom ASGI wrappers that need to forward lifespan events to mount_mcp()

[0.1.1] - 2026-03-09

Fixed

  • SCHEMA_PATH_PREFIX and EXCLUDE_PATHS — These settings were defined and documented but never implemented; paths are now filtered in get_processor() before schema processing
  • Streamable-HTTP test — Fixed incorrect test assertion that checked run() kwargs instead of server.settings attributes
  • Claude Code config path — Updated documentation to use ~/.claude.json instead of the incorrect ~/.claude/claude_code_config.json
  • Thread safety — Added double-checked locking with threading.Lock() to get_mcp_server() and get_processor() singletons, preventing race conditions under multi-worker deployments
  • Recursion depth limits — resolve_ref(), generate_example_from_schema(), and generate_example_value() now enforce a max depth of 10 to prevent stack overflow from circular $ref chains
  • Input validation — All MCP tool functions now validate path (must start with /) and method (must be a valid HTTP method) before processing
  • Code generation safety — Added _sanitize_identifier() and _sanitize_string_literal() to prevent code injection in generated JS/TS snippets from malicious schema values
  • servers[0]["url"] KeyError — Safely handle server entries missing the url key
  • Empty enum guard — generate_example_value() returns None instead of crashing on empty enum lists
  • Adapter error messages — Override adapter loading now provides descriptive ImportError messages when module or class is not found
  • Streamable-HTTP transport — Fixed TypeError in runmcpserver command when using --transport streamable-http; host and port are now set via server.settings instead of being passed as run() kwargs

Changed

  • Frozen dataclasses — All 8 schema dataclasses (Endpoint, Parameter, Response, etc.) are now immutable with frozen=True
  • Public API — Renamed _generate_example_value() to generate_example_value() (public method)
  • Removed format parameter — get_request_example() no longer accepts the unused format parameter
  • Removed default_app_config — Removed deprecated Django attribute (deprecated since Django 3.2; minimum supported version is 4.2)

Added

  • Python code generation — New requests (sync) and httpx (async) HTTP client generators with TypedDict definitions, type hints, and Google-style docstrings
  • Real TypeScript interfaces — Code snippets now generate actual TypeScript interfaces from OpenAPI schemas instead of placeholder types (RequestData, QueryParams)
  • JSDoc and docstrings — Generated code includes JSDoc (JS/TS) or Google-style docstrings (Python) with @param, @returns, @deprecated, @throws annotations
  • Usage examples — Generated snippets include commented usage examples with realistic data from the schema
  • Base URL from spec — Generated code pulls the base URL from the OpenAPI servers[0].url field
  • Auth method handling — Code generation now supports bearer, basic, and apiKey authentication based on actual security schemes (not just hardcoded Bearer)
  • Enriched JSON output — generate_code_snippet returns a metadata object with function name, endpoint info, auth details, parameter breakdown, and response summary
  • Auto-client selection — Automatically selects the appropriate HTTP client for the language (e.g., python + fetch → requests)
  • $ref caching — SchemaProcessor now caches resolved $ref pointers for improved performance on large schemas
  • yasg logging — Unknown parameter locations in Swagger 2.0 → OpenAPI 3.0 conversion now emit warnings via logging
  • Management command tests — New tests/test_management.py covering runmcpserver command
  • Concurrency tests — New tests/test_concurrency.py verifying thread-safe singleton behavior
  • Edge-case tests — Tests for circular refs, empty enums, missing server URLs, schemas with no paths/components, malformed operations, and frozen dataclass immutability
  • Path filtering tests — New tests/test_filtering.py covering SCHEMA_PATH_PREFIX and EXCLUDE_PATHS
  • Settings tests — New tests/test_settings.py covering get_setting() and get_all_settings()
  • ASGI mount tests — New tests/test_urls.py covering mount_mcp() routing

[0.1.0] - 2026-03-08

Added

  • Initial release
  • Schema adapter layer with support for drf-spectacular, drf-yasg, and DRF built-in
  • Auto-detection of installed schema generators
  • Schema processor with OpenAPI 3.x parsing, $ref resolution, and example generation
  • 6 MCP resources: api://overview, api://endpoints, api://endpoints/{method}/{path}, api://schemas, api://schemas/{name}, api://auth
  • 7 MCP tools: search_endpoints, get_endpoint_detail, get_request_example, get_response_example, generate_code_snippet, list_schemas, get_schema_detail
  • Code generation for fetch, axios, and ky in JavaScript and TypeScript
  • Django management command runmcpserver with stdio and streamable-http transport
  • ASGI mount helper for embedding MCP alongside Django
  • Configurable settings via DRF_MCP_DOCS dict
  • Full test suite (62 tests)