Skip to content

Add autonomous ASSERT MCP server - #332

Draft
Aaron Aspinwall (AaronAspinwall123) wants to merge 18 commits into
mainfrom
aaspinwall/asserMCP
Draft

Add autonomous ASSERT MCP server#332
Aaron Aspinwall (AaronAspinwall123) wants to merge 18 commits into
mainfrom
aaspinwall/asserMCP

Conversation

@AaronAspinwall123

@AaronAspinwall123 Aaron Aspinwall (AaronAspinwall123) commented Aug 25, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • add the release-grade MCP v2 server foundation and strict workspace containment
  • expose canonical config, pipeline, result, artifact, and library services through typed contracts
  • add 18 read-only inspect tools plus lazy schema, config, suite, run, transcript, and opaque-artifact resources
  • persist identity-validated per-suite run catalogs so large metadata-only listings avoid opening every run summary
  • add typed config validation, atomic ETag saves, and pure evaluation preflight with exact stage/cache plans and spend-policy enforcement
  • add explicit model-backed config design plus timeout-bounded target probing in an isolated subprocess with sanitation and process-tree cleanup
  • publish author/full capability contracts, policy limits, stable schemas, and legacy/modern protocol behavior

Validation

  • 1,346 passed, 28 skipped, and 474 subtests passed
  • 84 focused authoring, preflight, probe, cache, CLI, and MCP protocol tests passed
  • 100-suite / 1,000-run listing remains about 7.3 seconds against the existing 20-second gate
  • built-wheel MCP author-mode smoke passed

Remaining work

  • add persisted, idempotent job execution and truthful stage/case progress
  • add cancellation, startup recovery, retry, and process-tree escalation
  • add versioned curation, trace judging, and the remaining optional capability packs

Introduce the optional MCP server package, lazy CLI entry points, typed server discovery, capability modes, and in-memory plus stdio protocol coverage.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Persist lightweight run and suite summaries, build stale-aware JSONL indexes, and expose paginated result queries and comparisons through a shared repository used by the CLI.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Expose typed config validation, ETag saves, pure run planning, model-backed design, and isolated target probing through explicit MCP capability groups.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Integrate upstream scenario presets, permissibility metrics, and sandbox targets across the canonical config, strict MCP containment, preflight, probing, and discovery surfaces.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Add idempotent SQLite-backed jobs, isolated snapshot workers, bounded diagnostics, and start/list/get MCP tools with operator queue controls.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Bind persisted requests and reusable artifacts to immutable hashes, harden portable output identities and trace execution policy, and preserve curation and trace-result integrity across retries and recovery.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Increase the Tier 1 timeout so fresh dependency resolution does not consume the entire job budget before the repository test suite starts.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Run Tier 1 against the supported MCP v2 environment instead of combining it with Phoenix dependencies that still require MCP v1.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Assert that inference-only preflight exposes the exact immutable test-set version later bound into the persisted job request.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
Mark sanitized application failures as SDK ToolError instances so MCP 2.1 returns their stable model-correctable payload while continuing to redact unexpected crashes.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: cb4bac77-8743-43a4-8f2f-d1f7184cb965
@github-actions
github-actions Bot requested a review from tangym August 28, 2026 16:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant