- Created core TypeScript types for scenarios, annotated logs, and test results
- Implemented validation utilities with Zod schemas
- Added comprehensive unit tests for validation logic
- Set up TypeScript and test runner configuration using tsx
- Tests passing for scenario validation and log comparison utilities

- update_scenarios.md: Creates/updates scenarios based on MCP spec
- update_sdk.md: Generates SDK-specific test implementations
- update_goldens.md: Captures reference logs with TypeScript SDK
- cross_test_sdks.md: Runs cross-SDK compliance test matrix

- Created data.json with 3 test servers (CalcServer, FileServer, ErrorServer)
- Defined 25 test scenarios covering:
  - Basic tool invocation and elicitation flows
  - Multi-client interactions and per-client state
  - Resource management and subscriptions
  - Error handling, timeouts, and cancellation
  - All transport types (stdio, SSE, streamable HTTP)
  - Protocol features (progress, logging, roots, version negotiation)
- Added validation script to verify scenario structure
- All scenarios validated successfully

- Created StdioInterceptor class to capture JSON-RPC messages between processes
- Uses transform streams to intercept and log messages bidirectionally
- Annotates messages with sender, recipient, timestamp, and transport metadata
- Added unit tests verifying basic functionality
- Tests passing for stdio transport interception

- Created SSEInterceptor class as HTTP proxy for Server-Sent Events
- Intercepts and logs JSON-RPC messages sent via POST and SSE streams
- Forwards headers and maintains session state
- Added unit tests for server lifecycle and message capture
- Cleaned up unused imports in stdio interceptor
- All tests passing for both stdio and SSE interceptors

- Created StreamableHTTPInterceptor with full session management
- Handles POST, GET (SSE), and DELETE requests
- Tracks sessions and forwards appropriate headers
- Includes comprehensive metadata in logged messages
- Added unit tests for all major functionality
- All 15 interceptor tests passing (stdio, SSE, streamable HTTP)

- Implemented mcp-mitm CLI tool for man-in-the-middle logging
- Supports all three transports: stdio, SSE, streamable-http
- Logs annotated JSON-RPC messages to JSONL files
- Includes comprehensive CLI argument parsing and validation
- Added E2E tests verifying CLI functionality
- All 23 tests passing including interceptor and CLI tests

…Script SDK

- Added comprehensive server implementation with CalcServer, FileServer, and ErrorServer
- CalcServer features:
  - Basic arithmetic operations (add, ambiguous_add with elicitation)
  - Per-client trigonometric function management (cos/sin conditionally available)
  - Mutable resource management (special-number)
  - Sampling-based expression evaluation with progress tracking
  - Static prompt for mathematical problem solving
- FileServer features:
  - In-memory file system with write/delete operations
  - Static and templated resource access
  - Resource update notifications for subscribed files
  - Code review and file summarization prompts
- ErrorServer features:
  - Error testing scenarios (always_error, timeout, invalid_response)
  - Cancellation support for long-running operations
- All servers support stdio, SSE, and streamable-http transports
- Updated client.ts to use shared Scenarios type from compliance/src/types.ts

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

…omment

- Updated MITM CLI to accept --scenario-id flag
- Fetches scenario description from data.json
- Writes description as comment lines (// prefix) at start of log file
- Added tests for scenario ID functionality

- Added parseJSONLLog function that parses JSONL files with comment support
- Comments (lines starting with //) are extracted as description
- Added comprehensive tests for the new functionality
- Function returns both the parsed messages and the optional description

- Fixed path to use process.cwd() correctly when running from compliance directory
- Removed unused parseArgs import

- Created generate-goldens.ts script to run all scenarios and capture logs
- Added npm script 'generate-goldens' for easy execution
- Script supports stdio and SSE transports (streamable-http pending)
- Added --no-cache flag to test script to prevent tsx from generating artifacts

- Remove rootDir restriction to allow imports from parent directories
- Fix duplicate ListPromptsResult import
- Note: SDK still has API compatibility issues that need addressing

- Updated @modelcontextprotocol/sdk dependency from 1.0.0 to 1.15.0
- Fixed client.ts to use correct API for SDK 1.15.0:
  - Updated all client method calls to use object parameters instead of strings
  - Removed event handler methods that don't exist in SDK 1.15.0
  - Added comments for features not available in current SDK version
- Fixed server.ts to use correct API for SDK 1.15.0:
  - Removed extra parameter from tool/resource handlers
  - Fixed resource registration to use 4-parameter format
  - Changed prompt argumentSchema to argsSchema
  - Replaced system role with user/assistant roles in prompts
  - Simplified transport setup for SSE (not fully implemented)
- Updated test binaries to use relative paths with import.meta.url

The TypeScript SDK now builds successfully and binaries are functional.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

…ures

- Use SDK's built-in RegisteredTool.enable()/disable() methods
- Remove manual trigAllowed state checking in cos/sin implementations
- Automatic tool change events are now handled by the SDK
- Update comments to reflect current SDK capabilities

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add tsx command to mitm invocation in generate-goldens script
- Add allowUnknownOption() to test-client to handle mitm options
- Fix scenario data path resolution (remove duplicate 'compliance' in path)
- Generate initial golden files for scenarios 1-7

The mitm tool already had --log support but the test-client was
rejecting it as an unknown option. Now the full pipeline works
for stdio scenarios.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

… appending

- Modified generate-goldens to continue running all scenarios even when some fail
- Fixed mitm tool to replace log files instead of appending (was using 'a' flag, now uses 'w')
- Generated golden files for 20 out of 25 scenarios (5 failures due to missing features)

The failures are expected:
- Scenario 8: Pagination not implemented
- Scenario 14: SSE transport not implemented in test harness
- Scenario 15: Progress tracking not implemented
- Scenario 18: Prompt pagination not implemented
- Scenario 19: Streamable HTTP transport not implemented

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Replace process.cwd() with __dirname to find scenarios/data.json
- Remove failed environment variable approach
- Add --scenario-id flag back to mitm invocations
- Issue: --scenario-id is still being consumed before reaching mitm

- Add -- separator in generate-goldens to prevent test-client from parsing mitm's --scenario-id
- Remove .allowUnknownOption() from test-client since we now use -- to separate args
- Use __dirname instead of process.cwd() for reliable path resolution in mitm
- Change log file flags from 'a' to 'w' to replace content instead of appending
- Successfully regenerated 23 out of 25 golden files with scenario descriptions

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add CrossSDKRunner module to execute SDK client/server combinations
- Create test suite using Node's test runner with describe/it blocks
- Add comparison logic to validate captured traffic against golden files
- Create script to orchestrate cross-SDK testing with SDK selection
- Add npm script for easy cross-SDK test execution

The cross-SDK testing allows running any SDK's client against any SDK's
server and comparing the captured JSON-RPC traffic with golden files to
ensure protocol compliance across implementations.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Move compliance/goldens/ to compliance/scenarios/goldens/
- Update all references in generate-goldens.ts and cross-sdk.test.ts
- Better organization: keeps scenario data and golden files together

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

…er/recipient

- Remove timestamp field from AnnotatedJSONRPCMessage type
- Update all interceptors to stop adding timestamps to messages
- Update generate-goldens and cross-sdk-runner to pass client-id and server-id to MITM
- Remove timestamp normalization from validation code and tests
- Ensure proper sender/recipient info (e.g. client1, CalcServer) is included

This simplifies log comparison by removing non-deterministic timestamps and
ensures clear identification of message senders/recipients in the logs.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

…/to fields

- Move sender/recipient from metadata to top-level from/to fields
- Remove transport field from metadata (no longer needed)
- Place from/to fields before message for better readability
- Update all interceptors and validation code to use new structure
- Ensure metadata is optional and only contains streamable_http_metadata
- Update tests to match new structure

This simplifies the message structure and makes sender/recipient more
prominent in the logs.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

Implement elicitation request handler in the TypeScript SDK test client
to properly handle the ambiguous_add scenario (scenario 2). The handler
responds with value 20 when asked for 'b' parameter, allowing the
scenario to complete successfully.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>