Add action to generate and upload MCP-Eval badges from JSON report (#34)

## Summary

- Add a badge generation action and update workflow to use it.

## Checklist

- [ ] Tests added/updated
- [ ] Docs updated (README, docs/*.mdx)
- [x] Lint passes locally
- [ ] Linked issue (if any)

## Screenshots / Logs

If applicable, add screenshots or logs.

## Breaking Changes

- [ ] Yes
- [x] No

If yes, describe the migration path.

Author: StreetLamb

.github/actions/mcp-eval/badges/action.yaml

@@ -0,0 +1,101 @@
+name: "Generate MCP-Eval Badges"
+description: "Generate badges from MCP-Eval JSON report using shields.io"
+branding:
+  icon: award
+  color: green
+inputs:
+  report-path:
+    description: "Path to the MCP-Eval JSON report file"
+    required: true
+  output-dir:
+    description: "Directory to write generated badge files (optional, for caching)"
+    default: "mcpeval-reports/badges"
+    required: false
+  format:
+    description: "Output format: svg, endpoint, or both (default: both)"
+    default: "both"
+    required: false
+  tests-label:
+    description: "Label text for the tests badge"
+    default: "mcp-tests"
+    required: false
+  coverage-label:
+    description: "Label text for the coverage badge"
+    default: "mcp-cov"
+    required: false
+  upload-artifacts:
+    description: "Upload badges as workflow artifacts"
+    default: "false"
+    required: false
+  artifact-name:
+    description: "Name for the uploaded badge artifacts"
+    default: "mcpeval-badges"
+    required: false
+outputs:
+  tests-badge-path:
+    description: "Path to the generated tests badge SVG (if output-dir is set)"
+    value: ${{ steps.generate.outputs.tests_badge_path }}
+  coverage-badge-path:
+    description: "Path to the generated coverage badge SVG (if output-dir is set)"
+    value: ${{ steps.generate.outputs.coverage_badge_path }}
+runs:
+  using: "composite"
+  steps:
+    - name: Generate badges
+      id: generate
+      shell: bash
+      run: |
+        set -euo pipefail
+        
+        # Check if local script exists, otherwise fetch from upstream
+        if [ -f "scripts/generate_badges.py" ]; then
+          echo "Using local badge generation script"
+          SCRIPT_PATH="scripts/generate_badges.py"
+        else
+          echo "Fetching badge generation script from upstream"
+          # Create a temporary directory for the mcp-eval script
+          mkdir -p .mcp-eval-action
+          cd .mcp-eval-action
+          
+          # Initialize git and configure sparse checkout
+          git init
+          git remote add origin https://github.com/lastmile-ai/mcp-eval.git
+          git config core.sparseCheckout true
+          
+          # Configure sparse checkout to only get the script we need
+          echo "scripts/generate_badges.py" >> .git/info/sparse-checkout
+          
+          # Fetch and checkout the specific file (pinned to a stable commit)
+          # TODO: Update this to a specific tag/release when available
+          git fetch --depth=1 origin main
+          git checkout origin/main
+          
+          # Move back to the workspace root
+          cd ..
+          SCRIPT_PATH=".mcp-eval-action/scripts/generate_badges.py"
+        fi
+        
+        # Run the badge generation script with uv
+        uv run "$SCRIPT_PATH" \
+          --report "${{ inputs.report-path }}" \
+          --outdir "${{ inputs.output-dir }}" \
+          --label-tests "${{ inputs.tests-label }}" \
+          --label-cov "${{ inputs.coverage-label }}" \
+          --format "${{ inputs.format }}"
+        
+        # Set output paths if badges were generated
+        if [ -n "${{ inputs.output-dir }}" ]; then
+          if [ -f "${{ inputs.output-dir }}/tests.svg" ] && [ -f "${{ inputs.output-dir }}/coverage.svg" ]; then
+            echo "tests_badge_path=$(realpath ${{ inputs.output-dir }}/tests.svg)" >> $GITHUB_OUTPUT
+            echo "coverage_badge_path=$(realpath ${{ inputs.output-dir }}/coverage.svg)" >> $GITHUB_OUTPUT
+          fi
+        fi
+
+    - name: Upload badge artifacts
+      if: ${{ inputs.upload-artifacts == 'true' && inputs.output-dir != '' }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ inputs.artifact-name }}
+        path: ${{ inputs.output-dir }}
+        retention-days: 14
+        if-no-files-found: warn

.github/workflows/mcpeval-reusable.yml

@@ -43,6 +43,10 @@ on:
         required: false
         type: boolean
         default: false
+      deploy-pages-branch:
+        required: false
+        type: string
+        default: 'refs/heads/main'
     secrets:
       ANTHROPIC_API_KEY:
         required: false
@@ -53,6 +57,9 @@ jobs:
   run:
     name: Run MCP-Eval
     runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: write
     outputs:
       json: ${{ steps.mcpeval.outputs.results-json-path }}
       md: ${{ steps.mcpeval.outputs.results-md-path }}
@@ -63,7 +70,7 @@ jobs:
 
       - name: Run MCP-Eval
         id: mcpeval
-        uses: ./.github/actions/mcp-eval/run
+        uses: lastmile-ai/mcp-eval/.github/actions/mcp-eval/run@main
         with:
           python-version: ${{ inputs.python-version }}
           working-directory: ${{ inputs.working-directory }}
@@ -78,21 +85,49 @@ jobs:
           ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
 
-      - name: Generate badges
-        run: |
-          set -euo pipefail
-          uv run scripts/generate_badges.py --report "${{ steps.mcpeval.outputs.results-json-path }}" --outdir mcpeval-reports/badges
-
-      - name: Upload badge artifacts
-        uses: actions/upload-artifact@v4
+      - name: Generate and upload badges
+        uses: lastmile-ai/mcp-eval/.github/actions/mcp-eval/badges@main
         with:
-          name: mcpeval-badges
-          path: mcpeval-reports/badges
+          report-path: ${{ steps.mcpeval.outputs.results-json-path }}
+          output-dir: badges
+          format: 'both'
+          upload-artifacts: 'true'
+          artifact-name: mcpeval-badges
+      
+      # Post the Markdown report as a sticky PR comment for easy review
+      - name: Comment PR with MCP-Eval report
+        if: ${{ github.event_name == 'pull_request' }}
+        uses: actions/github-script@v7
+        env:
+          REPORT_PATH: ${{ steps.mcpeval.outputs.results-md-path }}
+        with:
+          script: |
+            const fs = require('fs');
+            const path = process.env.REPORT_PATH;
+            let body = '<!-- mcpeval-report -->\n';
+            body += '## MCP-Eval Report\n\n';
+            try {
+              const content = fs.readFileSync(path, 'utf8');
+              body += content;
+            } catch (e) {
+              body += '_No report found at ' + path + '_\n';
+            }
+            const { owner, repo } = context.repo;
+            const issue_number = context.issue.number;
+
+            // Find existing sticky comment
+            const { data: comments } = await github.rest.issues.listComments({ owner, repo, issue_number, per_page: 100 });
+            const previous = comments.find(c => c.user.type === 'Bot' && c.body.startsWith('<!-- mcpeval-report -->'));
+            if (previous) {
+              await github.rest.issues.updateComment({ owner, repo, comment_id: previous.id, body });
+            } else {
+              await github.rest.issues.createComment({ owner, repo, issue_number, body });
+            }
 
   pages:
-    name: Publish Report to Pages
+    name: Publish report and badges to GitHub Pages
     needs: run
-    if: ${{ inputs.deploy-pages }}
+    if: ${{ inputs.deploy-pages && github.event_name == 'push' && github.ref == inputs.deploy-pages-branch }}
     runs-on: ubuntu-latest
     permissions:
       pages: write
@@ -101,8 +136,20 @@ jobs:
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
+    concurrency:
+      group: "pages"
+      cancel-in-progress: false
     steps:
-      - name: Download artifacts
+      - name: Checkout
+        uses: actions/checkout@v4
+      
+      - name: Download badge artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: mcpeval-badges
+          path: badges/
+
+      - name: Download report artifacts
         uses: actions/download-artifact@v4
         with:
           name: ${{ inputs.artifact-name }}
@@ -112,13 +159,23 @@ jobs:
         run: |
           set -euo pipefail
           mkdir -p site
+          
+          # Copy badges to site
+          if [[ -d badges ]]; then
+            cp -r badges site/
+          fi
+          
+          # Copy HTML report as index.html
           if [[ -f "mcpeval-artifacts/${{ inputs.reports-dir }}/${{ inputs.html-report }}" ]]; then
             cp mcpeval-artifacts/${{ inputs.reports-dir }}/${{ inputs.html-report }} site/index.html
           else
             file=$(find mcpeval-artifacts -name "*.html" | head -n 1 || true)
             if [[ -n "$file" ]]; then cp "$file" site/index.html; else echo '<h1>No report available</h1>' > site/index.html; fi
           fi
 
+      - name: Setup Pages
+        uses: actions/configure-pages@v5
+
       - name: Upload Pages artifact
         uses: actions/upload-pages-artifact@v3
         with:

.github/workflows/mcpeval.yml

@@ -10,129 +10,15 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
-permissions:
-  contents: read
-  pull-requests: write
-
 jobs:
-  tests:
-    name: Run MCP-Eval
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        shell: bash
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      # Tip: set your LLM provider keys in repo/org secrets
-      # Settings discovery follows mcp-agent/mcpeval configs in your repo.
-      - name: Run MCP-Eval (uv)
-        id: mcpeval
-        uses: ./.github/actions/mcp-eval/run
-        with:
-          python-version: "3.11"
-          working-directory: .
-          run-args: "-v"
-          tests: tests/
-          reports-dir: mcpeval-reports
-          json-report: mcpeval-results.json
-          markdown-report: mcpeval-results.md
-          html-report: mcpeval-results.html
-          artifact-name: mcpeval-artifacts
-          pr-comment: "true"
-          set-summary: "true"
-          upload-artifacts: "true"
-        env:
-          # Provide at least one provider key; both are supported
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-
-      - name: Generate badges
-        run: |
-          set -euo pipefail
-          uv run scripts/generate_badges.py --report "${{ steps.mcpeval.outputs.results-json-path }}" --outdir mcpeval-reports/badges
-
-      - name: Upload badge artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: mcpeval-badges
-          path: mcpeval-reports/badges
-
-      # Post the Markdown report as a sticky PR comment for easy review
-      - name: Comment PR with MCP-Eval report
-        if: ${{ github.event_name == 'pull_request' }}
-        uses: actions/github-script@v7
-        env:
-          REPORT_PATH: ${{ steps.mcpeval.outputs.results-md-path }}
-        with:
-          script: |
-            const fs = require('fs');
-            const path = process.env.REPORT_PATH;
-            let body = '<!-- mcpeval-report -->\n';
-            body += '## MCP-Eval Report\n\n';
-            try {
-              const content = fs.readFileSync(path, 'utf8');
-              body += content;
-            } catch (e) {
-              body += '_No report found at ' + path + '_\n';
-            }
-            const { owner, repo } = context.repo;
-            const issue_number = context.issue.number;
-
-            // Find existing sticky comment
-            const { data: comments } = await github.rest.issues.listComments({ owner, repo, issue_number, per_page: 100 });
-            const previous = comments.find(c => c.user.type === 'Bot' && c.body.startsWith('<!-- mcpeval-report -->'));
-            if (previous) {
-              await github.rest.issues.updateComment({ owner, repo, comment_id: previous.id, body });
-            } else {
-              await github.rest.issues.createComment({ owner, repo, issue_number, body });
-            }
-
-  # Optional: Publish the HTML report to GitHub Pages on main/master pushes
-  pages:
-    name: Publish Report to Pages
-    needs: tests
-    if: ${{ github.event_name == 'push' && (github.ref == 'refs/heads/main' || github.ref == 'refs/heads/master') }}
-    runs-on: ubuntu-latest
+  call-mcpeval:
+    uses: lastmile-ai/mcp-eval/.github/workflows/mcpeval-reuseable.yml@main
+    with:
+      deploy-pages: true
+      tests: examples/mcp_server_fetch/tests/test_simple_decorator.py
     permissions:
+      contents: read
       pages: write
       id-token: write
-      contents: read
-    environment:
-      name: github-pages
-      url: ${{ steps.deployment.outputs.page_url }}
-    steps:
-      - name: Download artifacts
-        uses: actions/download-artifact@v4
-        with:
-          name: mcpeval-artifacts
-          path: ./mcpeval-artifacts
-
-      - name: Prepare site
-        run: |
-          set -euo pipefail
-          mkdir -p site
-          # Prefer the configured HTML filename; fallback to first HTML we find
-          if [[ -f "mcpeval-artifacts/mcpeval-reports/mcpeval-results.html" ]]; then
-            cp mcpeval-artifacts/mcpeval-reports/mcpeval-results.html site/index.html
-          else
-            file=$(find mcpeval-artifacts -name "*.html" | head -n 1 || true)
-            if [[ -n "$file" ]]; then cp "$file" site/index.html; else echo '<h1>No report available</h1>' > site/index.html; fi
-          fi
-          # Include badges if available
-          if [[ -d "mcpeval-artifacts/mcpeval-reports/badges" ]]; then
-            mkdir -p site/badges
-            cp -r mcpeval-artifacts/mcpeval-reports/badges/*.svg site/badges/ || true
-          fi
-
-      - name: Upload Pages artifact
-        uses: actions/upload-pages-artifact@v3
-        with:
-          path: ./site
-
-      - name: Deploy to GitHub Pages
-        id: deployment
-        uses: actions/deploy-pages@v4
-
-
+      pull-requests: write
+    secrets: inherit