v0.4.0

Added

• Expose qe_run_openai_compat_agent as an experiment-only, env-gated standalone MCP runner for OpenAI-compatible endpoints.
• Add server wiring, schema/help/list dispatch, smoke coverage, and MCP stdio wiring tests.

Changed

• Keep qe_delegate_agent target engines limited to claude/codex while documenting openai-compat as standalone.
• Update README migration notes and tool surface documentation for 0.4.0.

v0.2.4

• Add cross-platform CI for qe-mcp on Linux, macOS, and Windows with Node 20/22.\n- Fix Claude stdio compatibility and registry sync behavior from the v0.2.3 line.\n- Fix Windows maintenance job execution by avoiding undefined env values and running npm jobs through Node/npm CLI.

v0.2.3

• Support Claude Code newline JSON-RPC stdio framing while keeping Codex Content-Length framing.\n- Add Claude MCP sync type=stdio so Claude Code user config health checks can connect.\n- Make qe-mcp init-registry idempotent by upserting qeExpertLibrary.\n- Add selftest coverage for Claude framing, registry idempotency, and Windows-style client config paths.

v0.2.2

• Reduce default MCP initialize capabilities to tools only to avoid client startup stalls.\n- Keep optional resources/prompts behind explicit environment flags.\n- Add selftest coverage for the startup capability payload.\n- Bump serverInfo/package version to 0.2.2.

qe-mcp v0.2.1

Published @inho-team/qe-mcp@0.2.1 to npm.

Changes:

• Bump MCP server/package version to 0.2.1.
• Normalize npm bin paths before publish.

Verification:

• npm run check
• npm run selftest
• npm run runner:smoke
• npm view @inho-team/qe-mcp version => 0.2.1

qe-mcp v0.2.0

v0.2.0

Adds opt-in cross-agent runner tools to the QE MCP server.

Added

• qe_run_codex_agent for bounded local Codex CLI execution.
• qe_run_claude_agent for bounded local Claude CLI execution.
• qe_cross_agent_help for passive local runner capability discovery.
• Runner subprocess lifecycle handling, timeout cleanup, output normalization, environment filtering, cwd policy, recursion guard, and active-runner concurrency guard.
• Selftest coverage for existing expert-library tools and new runner contracts.

Notes

• Active runners are default read-only/plan-mode and return normalized AgentRunResult payloads.
• npm publish was not performed in this environment because npm auth is unavailable.

v0.1.0

QE MCP v0.1.0

Initial standalone QE expert-library MCP companion release.

Verification:

• npm run check
• npm run selftest