jev-mcp-python 0.1.0__tar.gz

jev_mcp_python-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,98 @@
+name: ci
+
+on:
+  push:
+  pull_request:
+
+permissions:
+  contents: read
+
+# ruff -> pyright -> unit -> property -> contract -> parity -> security -> build -> MCP smoke (ROADMAP CI).
+jobs:
+  ruff:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make lint
+
+  pyright:
+    needs: ruff
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make typecheck
+
+  unit:
+    needs: pyright
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make unit
+
+  property:
+    needs: unit
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make property
+      - run: make policy-coverage
+
+  contract:
+    needs: property
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make contract
+
+  parity:
+    needs: contract
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      # Node 24 for fixture replay and the serializer differential tests (P0/P2).
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+      - run: uv sync --locked --all-extras
+      - run: make parity
+
+  security:
+    needs: parity
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make security
+
+  build:
+    needs: security
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: make build
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  smoke:
+    needs: build
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+      - run: uv sync --locked --all-extras
+      - run: make smoke

jev_mcp_python-0.1.0/.github/workflows/release.yml

@@ -0,0 +1,158 @@
+name: Release
+
+# Conventional Commits on main -> a release pull request. Merging it -> a draft GitHub Release with its
+# tag (release-please-config.json: draft + force-tag-creation). The tagged commit is then built, its
+# assets and checksums uploaded, re-downloaded, installed into a fresh venv, and checked; only then is
+# the release published, and the verified wheel and sdist are uploaded to PyPI through trusted
+# publishing (OIDC, no stored token). Any failure leaves the release a draft.
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+permissions: {}
+
+# One release at a time; a queued run waits instead of cancelling one mid-release.
+concurrency:
+  group: release
+  cancel-in-progress: false
+
+defaults:
+  run:
+    shell: bash
+
+jobs:
+  version:
+    name: Version
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write # release commits, tags, draft releases
+      pull-requests: write # the release pull request
+      issues: write # its autorelease labels (the action's documented minimum)
+    outputs:
+      release_created: ${{ steps.releaser.outputs.release_created }}
+      tag_name: ${{ steps.releaser.outputs.tag_name }}
+      sha: ${{ steps.releaser.outputs.sha }}
+    steps:
+      # Config and manifest only: a `release-type` input would bypass both and drop `draft`.
+      - id: releaser
+        uses: googleapis/release-please-action@45996ed1f6d02564a971a2fa1b5860e934307cf7 # v5.0.0
+        with:
+          token: ${{ github.token }}
+          config-file: release-please-config.json
+          manifest-file: .release-please-manifest.json
+
+  build:
+    name: Build
+    needs: version
+    if: needs.version.outputs.release_created == 'true'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write # upload assets onto the draft release
+    env:
+      TAG: ${{ needs.version.outputs.tag_name }}
+      TAG_SHA: ${{ needs.version.outputs.sha }}
+      GH_TOKEN: ${{ github.token }}
+      GH_REPO: ${{ github.repository }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          ref: refs/tags/${{ needs.version.outputs.tag_name }}
+          persist-credentials: false
+      - name: Confirm the tagged commit and a draft release
+        run: |
+          set -euo pipefail
+          test "$(git rev-parse HEAD)" = "$TAG_SHA"
+          test "$(gh release view "$TAG" --json isDraft --jq .isDraft)" = true
+      - uses: astral-sh/setup-uv@d0cc045d04ccac9d8b7881df0226f9e82c39688e # v6.8.0
+        with:
+          enable-cache: false
+      - name: Build one wheel and one sdist at the tag version
+        run: |
+          set -euo pipefail
+          rm -rf dist
+          uv build
+          uv run --no-project scripts/check_release_version.py dist dist "$TAG"
+      - name: Write checksums
+        working-directory: dist
+        run: |
+          set -euo pipefail
+          sha256sum -- *.whl *.tar.gz > SHA256SUMS.txt
+          cat SHA256SUMS.txt
+      - name: Upload onto the draft release
+        run: |
+          set -euo pipefail
+          gh release upload "$TAG" dist/*.whl dist/*.tar.gz dist/SHA256SUMS.txt --clobber
+
+  verify:
+    name: Verify and publish
+    needs: [version, build]
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write # drafts and their assets are visible only with write; also the final undraft
+    env:
+      TAG: ${{ needs.version.outputs.tag_name }}
+      GH_TOKEN: ${{ github.token }}
+      GH_REPO: ${{ github.repository }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          ref: refs/tags/${{ needs.version.outputs.tag_name }}
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@d0cc045d04ccac9d8b7881df0226f9e82c39688e # v6.8.0
+        with:
+          enable-cache: false
+      - name: Download the release assets and check their sums
+        run: |
+          set -euo pipefail
+          assets="$RUNNER_TEMP/assets"
+          gh release download "$TAG" --dir "$assets"
+          cd "$assets"
+          test "$(wc -l < SHA256SUMS.txt)" -eq 2
+          sha256sum --check --strict SHA256SUMS.txt
+          uv run --no-project "$GITHUB_WORKSPACE/scripts/check_release_version.py" dist . "$TAG"
+      - name: Install the wheel into a fresh venv and check the installed version
+        run: |
+          set -euo pipefail
+          venv="$RUNNER_TEMP/verify-venv"
+          uv venv --python "$(cat .python-version)" "$venv"
+          uv pip install --python "$venv/bin/python" "$RUNNER_TEMP"/assets/*.whl
+          # -I: no checkout, script directory, or PYTHONPATH on sys.path; jev_mcp must come from the wheel.
+          cd "$RUNNER_TEMP"
+          "$venv/bin/python" -I "$GITHUB_WORKSPACE/scripts/check_release_version.py" installed "$TAG"
+      - name: Publish the release
+        run: |
+          set -euo pipefail
+          test "$(gh release view "$TAG" --json isDraft --jq .isDraft)" = true
+          gh release edit "$TAG" --draft=false
+
+  pypi:
+    name: Publish to PyPI
+    needs: [version, build, verify]
+    if: needs.version.outputs.release_created == 'true'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write # PyPI trusted publishing (OIDC); no stored token
+    environment: pypi
+    env:
+      TAG: ${{ needs.version.outputs.tag_name }}
+      GH_TOKEN: ${{ github.token }}
+      GH_REPO: ${{ github.repository }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          ref: refs/tags/${{ needs.version.outputs.tag_name }}
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@d0cc045d04ccac9d8b7881df0226f9e82c39688e # v6.8.0
+        with:
+          enable-cache: false
+      - name: Download the verified wheel and sdist from the release
+        run: |
+          set -euo pipefail
+          assets="$RUNNER_TEMP/assets"
+          gh release download "$TAG" --dir "$assets"
+          test "$(ls "$assets"/*.whl "$assets"/*.tar.gz | wc -l)" -eq 2
+      - name: Upload to PyPI through trusted publishing
+        run: uv publish --trusted-publishing always --no-attestations "$RUNNER_TEMP/assets"

jev_mcp_python-0.1.0/.gitignore

@@ -0,0 +1,23 @@
+*.pyc
+__pycache__/
+.venv/
+dist/
+.coverage
+coverage.xml
+.pytest_cache/
+.ruff_cache/
+.hypothesis/
+.benchmarks/
+tests/parity/reference/
+docs/architecture.visual-check.*
+evals/reports/*
+!evals/reports/.gitkeep
+!evals/reports/p8-pilot.md
+!evals/reports/bench150-sample.html
+!evals/reports/bench150.html
+!evals/reports/bench150.md
+.DS_Store
+.treehouse/
+AGENTS.md
+CLAUDE.md
+improvment.md

jev_mcp_python-0.1.0/.python-version

@@ -0,0 +1 @@
+3.12

jev_mcp_python-0.1.0/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.1.0"
+}

jev_mcp_python-0.1.0/CHANGELOG.md

@@ -0,0 +1,8 @@
+# Changelog
+
+## 0.1.0 (2026-09-22)
+
+
+### Features
+
+* initial release of jev-mcp-python 0.1.0 ([c31bae5](https://github.com/PyModel/jev-mcp/commit/c31bae5bdbcc928be99dfb908b2dafcfb3b11179))

jev_mcp_python-0.1.0/CONTEXT.md

@@ -0,0 +1,125 @@
+# Jev MCP
+
+An MCP server that turns TypeSafe's Jev model into purpose-built judgment tools for agents. Jev returns typed probabilities, not text; this server designs the questions and turns the answers into actions an agent can safely act on.
+
+## Language
+
+### Judgments
+
+**Jev**:
+TypeSafe's System One model: it answers typed questions about a state with probability distributions instead of generated text.
+_Avoid_: the LLM, the model (when the host agent's model could be meant)
+
+**State**:
+The observed material a judgment is made about (claims, evidence, candidates, a diff). State is evidence to evaluate, never instructions to follow.
+_Avoid_: context, prompt, input
+
+**Question**:
+One narrow typed ask about the state, of kind Choice, Noul, or Score. Questions in one request cannot see each other's answers.
+_Avoid_: prompt, query (query is a find/rerank input field)
+
+**Choice**:
+A question whose answer is one option from a named set, with a probability for every option.
+_Avoid_: enum, select, multiple-choice
+
+**Noul**:
+A question whose answer is the probability that a yes/no condition holds. A value near 0.5 means uncertain, not "medium".
+_Avoid_: boolean (Vercel's wire name for it), yes/no
+
+**Score**:
+A question whose answer is a probability-weighted position on an ordered list of concrete levels, 0-indexed.
+_Avoid_: rating, grade
+
+**Answer**:
+Jev's raw reply to one question, before validation.
+_Avoid_: result, response (response means the whole provider reply)
+
+**Confidence**:
+How concentrated an answer's distribution is. It is not correctness, and unknown confidence never satisfies a threshold.
+_Avoid_: certainty, accuracy
+
+**Margin**:
+The gap between the top and runner-up probabilities of a Choice answer.
+_Avoid_: lead, spread
+
+### Deciding
+
+**Action**:
+What the caller may do with a judgment: `auto` (stands alone), `review` (a human or stronger check should confirm), or `escalate` (must not proceed as is).
+_Avoid_: decision, status, verdict
+
+**Verdict**:
+A tool's semantic conclusion about one item, such as verified / contradicted / unsupported or answered / partial / absent. A verdict is not an action.
+_Avoid_: label, result
+
+**Policy**:
+The deterministic rules that turn validated answers and thresholds into actions. The model judges; policy decides.
+_Avoid_: rules engine, heuristics
+
+**Threshold**:
+A caller-tunable probability or confidence boundary a policy compares against, such as auto_accept or review_at.
+_Avoid_: cutoff, limit (limit means a Cap)
+
+**Fail Closed**:
+The rule that a missing, malformed, or out-of-bounds answer produces `invalid_response` and a non-auto action, never a default verdict.
+_Avoid_: fallback, graceful degradation
+
+**Truncated Context**:
+A document the judgment is made over (a doc, diff, claim, or evidence item) that exceeded a cap and was cut before reaching Jev. A judgment over truncated context can never be `auto`. Cut item text (a jev_find or jev_rerank candidate, a jev_classify item or class description) is not Truncated Context: it is reported to telemetry only and does not change the action. jev_extract's capped candidate universe is a separate rule of its own field decision.
+_Avoid_: partial input, clipped
+
+**Reason Code**:
+A stable machine-readable token explaining why a gate reached its action, such as `claims_contradicted`.
+_Avoid_: error code, message
+
+**Escape Hatch**:
+An extra Choice option (ask_user, investigate, none) that lets Jev decline to pick among supplied candidates.
+_Avoid_: abstain option, fallback option
+
+**Composite**:
+The weighted 0–1 summary of a patch review's four rubric scores.
+_Avoid_: overall score, quality score
+
+### Inputs
+
+**Claim**:
+An assertion to be checked against evidence. A claim is never evidence for itself or for the patch it describes.
+_Avoid_: statement, assertion (in field names)
+
+**Evidence**:
+The only material a claim may be judged against.
+_Avoid_: sources, context, proof
+
+**Candidate**:
+One caller-supplied option to be found, ranked, picked, or extracted. In jev_extract a candidate is a verbatim regex match, and the returned value is always one of them or null.
+_Avoid_: option (wire-level term), document, hit
+
+**Cap**:
+A hard size bound on an input; exceeding it either rejects the call or truncates, depending on the field.
+_Avoid_: limit, quota
+
+### Providers and parity
+
+**Provider**:
+A transport that delivers questions to Jev and returns answers: TypeSafe direct, OpenRouter, Cloudflare, Vercel, or a Jev-compatible endpoint.
+_Avoid_: backend, adapter, client
+
+**Envelope**:
+The provider response wrapper around the answers (answers object, usage, model). Providers validate only the envelope.
+_Avoid_: payload, body
+
+**Reference Implementation**:
+The TypeScript reference server 0.5.0 whose observable behavior the Python server must match.
+_Avoid_: the old server, the original, upstream
+
+**Parity**:
+Identical tool output from the Reference Implementation and the Python server given identical inputs and identical simulated answers.
+_Avoid_: compatibility, equivalence
+
+**Sanctioned Divergence**:
+An intentional, ADR-recorded difference from the Reference Implementation that the parity suite expects rather than flags.
+_Avoid_: known difference, deviation, bug-compat
+
+**Reference Quirk**:
+A behavior of the Reference Implementation that looks accidental and needs an explicit keep-or-diverge decision.
+_Avoid_: bug, legacy behavior

jev_mcp_python-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,82 @@
+# Contributing
+
+jev-mcp-python is a Python MCP server exposing TypeSafe's Jev judgment tools,
+frozen against a TypeScript reference server. The public repository is
+https://github.com/PyModel/jev-mcp. Open a pull request against `main`.
+
+## Read first
+
+- `README.md`: what the server is, how to run it, and the recorded eval numbers.
+- `CONTEXT.md`: the project vocabulary.
+- `ROADMAP.md`: phases and acceptance criteria.
+- `docs/adr/`: the decisions. The spec is frozen: a change the ADRs do not cover
+  needs a new ADR, not a silent design choice.
+
+## Set up
+
+Requires Python 3.12+ and [uv](https://docs.astral.sh/uv/).
+
+```sh
+uv sync --all-extras
+```
+
+## Prove a change
+
+Run every CI stage locally, with `TYPESAFE_API_KEY` unset:
+
+```sh
+env -u TYPESAFE_API_KEY make ci
+```
+
+`make ci` must pass. The parity stage needs Node 24 and skips locally without it.
+Report the command you ran and its result with the change.
+
+- Fix a bug by first adding a test that fails, then making it pass.
+- Never hand-edit `tests/parity/fixtures/`; change `tests/parity/cases/` and
+  re-record with `make parity-record`.
+- Never delete or weaken a test to make the build pass.
+
+## Paid gates are not part of a contribution
+
+`make security-live`, `make eval-live`, and `make ab` call the real TypeSafe API
+and spend money. A normal contribution does not run them and must not require
+them. They are run by the maintainer, within their budgets, when a change touches
+live behavior.
+
+## Offer a contribution
+
+1. Branch from the current `main`.
+2. Keep the change focused; one concern per commit, imperative subject line.
+3. Rebase onto `main` so the branch fast-forwards.
+4. Hand the branch to the maintainer with the `make ci` result.
+
+## Releases
+
+The Release workflow (`.github/workflows/release.yml`) cuts releases from
+Conventional Commits on `main`; there is no release trailer or manual tag. It
+keeps a release pull request open with the next version and changelog; merging
+it creates a draft GitHub Release and its `v<version>` tag. The workflow builds
+that tag, uploads the wheel, sdist, and `SHA256SUMS.txt`, installs the downloaded
+wheel into a fresh venv, and publishes the release only if the installed version
+matches the tag. A failed check leaves the release a draft. The verified wheel and
+sdist are then uploaded to PyPI through trusted publishing: no token is stored in
+the repository; the PyPI project (`jev-mcp-python`) must have a pending or active
+publisher for `PyModel/jev-mcp`, workflow `release.yml`, environment `pypi`.
+
+- The first release is `v0.1.0`: `pyproject.toml` already declares `0.1.0`,
+  `initial-version` in `release-please-config.json` records it, and
+  `.release-please-manifest.json` stays `0.0.0` (nothing released) until the
+  first release pull request is merged. That pull request needs a `feat` or
+  `fix` commit in the scanned history; the history starts at the initial commit.
+- The release pull request is opened with the workflow token, so GitHub does not
+  run `ci` on it. Run `env -u TYPESAFE_API_KEY make ci` on its branch before
+  merging.
+- The version gate is `scripts/check_release_version.py`, tested offline in
+  `tests/unit/test_release_version.py`.
+
+## License
+
+The copyright holder is Mohamed Elkholy.
+
+By contributing, you agree that your contributions are licensed under the MIT
+License in `LICENSE`.

jev_mcp_python-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Mohamed Elkholy⁠​‌​​‌‌​‌​‌‌​‌‌‌‌​‌‌​‌​​​​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​‌​‌​‌‌​​‌​​​​‌​​​​​​‌​​​‌​‌​‌‌​‌‌​​​‌‌​‌​‌‌​‌‌​‌​​​​‌‌​‌‌‌‌​‌‌​‌‌​​​‌‌‌‌​​‌
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

jev_mcp_python-0.1.0/Makefile

@@ -0,0 +1,93 @@
+# CI stages, in ROADMAP order. Every declared stage is real: pytest exits 5 (nothing collected)
+# when its directory is missing or empty, so a stage cannot pass vacuously. A new stage joins `ci`
+# in the same commit that adds its tests.
+PYTEST := uv run pytest
+
+.PHONY: ci lint typecheck unit property policy-coverage contract parity security build smoke eval eval-live security-live ab load
+
+ci: lint typecheck unit property policy-coverage contract parity security build smoke
+
+lint:
+	uv run ruff check
+	uv run ruff format --check
+
+typecheck:
+	uv run pyright
+
+unit:
+	$(PYTEST) tests/unit --cov=jev_mcp --cov-report=term-missing
+
+property:
+	$(PYTEST) tests/property
+
+# ROADMAP P3: the policy package keeps 100% branch coverage from its unit and property tests alone.
+policy-coverage:
+	$(PYTEST) tests/unit tests/property -q --cov=jev_mcp.policy --cov-branch --cov-fail-under=100 --cov-report=term-missing
+
+contract:
+	$(PYTEST) tests/contract
+
+parity:
+	$(PYTEST) tests/parity
+
+# ROADMAP P6: the scripted adversary, offline. Real now, so an empty stage fails (pytest exit 5).
+security:
+	$(PYTEST) tests/security
+
+# Live P6 gate against TypeSafe (paid, bounded; tests/security/test_live_typesafe.py). Fails without
+# TYPESAFE_API_KEY. Not part of `make ci` or the CI security job.
+security-live:
+	$(PYTEST) tests/security -m live
+
+build:
+	uv build
+
+# The integration tests, then the slow `uvx --from .` smoke test. A command-line `-m` replaces the addopts
+# one, so the smoke line selects `smoke` alone and never re-admits `live` (ADR-0027).
+smoke:
+	$(PYTEST) tests/integration
+	$(PYTEST) tests/integration -m smoke
+
+# ROADMAP P9: local overhead at 1/4/16/32/64 concurrent calls, provider stubbed (tests/load/). Timing-based,
+# so it stays out of `make ci`.
+load:
+	$(PYTEST) tests/load -m load -s
+
+# ROADMAP P7: offline eval scorer and calibration tests. No network, no API key.
+eval:
+	$(PYTEST) tests/evals
+
+# Live L3 smoke against TypeSafe (paid, bounded; evals/README.md). Fails without TYPESAFE_API_KEY.
+eval-live:
+	$(PYTEST) tests/evals -m live
+
+# ROADMAP P8 agent A/B pilot (paid, capped at 9 runs and 25 USD; evals/README.md § L4). Refuses unless
+# the caller passes JEV_AB_LIVE=1 on the command line; this target never sets it. Refuses to overwrite a
+# committed evals/reports/p8-pilot.md when neither its recorded runs nor a ledger are present.
+ab:
+	uv run python -m evals.ab.pilot
+
+# Parity fixtures (ROADMAP P0). The pinned TypeScript reference is fetched into
+# tests/parity/reference/ (gitignored) and run under Node 24.19.0.
+
+NODE ?= $(HOME)/.nvm/versions/node/v24.19.0/bin/node
+NPM ?= $(dir $(NODE))npm
+PARITY := tests/parity
+REFERENCE := $(PARITY)/reference
+
+.PHONY: parity-reference parity-record parity-verify
+
+# Fetch and build the reference at the pinned commit. Needs the network once;
+# later runs reuse the checkout and are offline.
+parity-reference:
+	PATH="$(dir $(NODE)):$$PATH" NODE="$(NODE)" NPM="$(NPM)" $(PARITY)/harness/reference.sh $(REFERENCE)
+
+# Regenerate tests/parity/fixtures/ from the pinned reference. Offline: the only
+# provider is the local fake in harness/lib.mjs.
+parity-record: parity-reference
+	PATH="$(dir $(NODE)):$$PATH" $(NODE) $(PARITY)/harness/record.mjs
+
+# Replay every committed fixture against the reference through the fake and
+# require byte-identical tool results.
+parity-verify: parity-reference
+	PATH="$(dir $(NODE)):$$PATH" $(NODE) $(PARITY)/harness/replay.mjs

jev_mcp_python-0.1.0/PKG-INFO

@@ -0,0 +1,18 @@
+Metadata-Version: 2.5
+Name: jev-mcp-python
+Version: 0.1.0
+Summary: Behavior-compatible Python MCP server exposing TypeSafe's Jev judgment tools.
+Author: Mohamed Elkholy
+License-Expression: MIT
+License-File: LICENSE
+Classifier: License :: OSI Approved :: MIT License
+Requires-Python: >=3.12
+Requires-Dist: httpx
+Requires-Dist: mcp<3,>=2.2
+Requires-Dist: pydantic-settings
+Requires-Dist: pydantic>=2
+Provides-Extra: telemetry
+Requires-Dist: opentelemetry-sdk; extra == 'telemetry'
+Requires-Dist: prometheus-client; extra == 'telemetry'
+Provides-Extra: typesafe
+Requires-Dist: typesafe-sdk; extra == 'typesafe'