Add MCP server endpoint for AI agent doc access

.gitignore

@@ -15,3 +15,4 @@ docs/**/*
 *.zip
 assets/stylesheets/components/_environment.scss
 assets/stylesheets/global/_icons.scss
+.mcp.json

lib/app.rb

@@ -105,6 +105,7 @@ class App < Sinatra::Application
 
   configure :test do
     set :docs_manifest_path, File.join(root, 'test', 'files', 'docs.json')
+    set :docs_path, File.join(root, 'test', 'files', 'docs')
   end
 
   def self.parse_docs
@@ -275,6 +276,14 @@ def service_worker_cache_name
     200
   end
 
+  require 'mcp/server'
+
+  post '/mcp' do
+    content_type :json
+    payload = JSON.parse(request.body.read)
+    Mcp::Server.handle(payload, settings).to_json
+  end
+
   %w(docs.json application.js application.css).each do |asset|
     class_eval <<-CODE, __FILE__, __LINE__ + 1
       get '/#{asset}' do

lib/mcp/server.rb

@@ -0,0 +1,169 @@
+module Mcp
+  # Dispatches a single JSON-RPC 2.0 request (already parsed into a Hash with
+  # string keys) to the appropriate MCP handler and returns a response Hash
+  # ready to be serialized back to the client.
+  module Server
+    TOOLS = [
+      {
+        'name' => 'devdocs_list_docsets',
+        'description' => 'List documentation sets available on this DevDocs instance. Returns paginated results with optional filtering.',
+        'inputSchema' => {
+          'type' => 'object',
+          'properties' => {
+            'offset' => { 'type' => 'integer', 'description' => 'Number of results to skip (default: 0)', 'minimum' => 0 },
+            'limit' => { 'type' => 'integer', 'description' => 'Maximum results to return (default: 50, max: 500)', 'minimum' => 1, 'maximum' => 500 },
+            'query' => { 'type' => 'string', 'description' => 'Filter by slug or name (case-insensitive substring match)' },
+          },
+          'additionalProperties' => false,
+        },
+      },
+      {
+        'name' => 'devdocs_search',
+        'description' => 'Search entry names/paths within one downloaded DevDocs doc set.',
+        'inputSchema' => {
+          'type' => 'object',
+          'properties' => {
+            'slug' => { 'type' => 'string' },
+            'query' => { 'type' => 'string' },
+          },
+          'required' => %w(slug query),
+          'additionalProperties' => false,
+        },
+      },
+      {
+        'name' => 'devdocs_get_page',
+        'description' => 'Fetch one entry from a DevDocs doc set as plain text.',
+        'inputSchema' => {
+          'type' => 'object',
+          'properties' => {
+            'slug' => { 'type' => 'string' },
+            'path' => { 'type' => 'string' },
+          },
+          'required' => %w(slug path),
+          'additionalProperties' => false,
+        },
+      },
+    ].freeze
+
+    def self.handle(request, app_settings)
+      case request['method']
+      when 'initialize'
+        respond(request, {
+          'protocolVersion' => '2024-11-05',
+          'capabilities' => { 'tools' => {} },
+          'serverInfo' => { 'name' => 'devdocs-mcp', 'version' => '1.0.0' },
+        })
+      when 'tools/list'
+        respond(request, { 'tools' => TOOLS })
+      when 'tools/call'
+        call_tool(request, app_settings)
+      else
+        error(request, -32601, "Unsupported method: #{request['method']}")
+      end
+    end
+
+    def self.error(request, code, message)
+      { 'jsonrpc' => '2.0', 'id' => request['id'], 'error' => { 'code' => code, 'message' => message } }
+    end
+
+    def self.call_tool(request, app_settings)
+      params = request['params']
+      case params['name']
+      when 'devdocs_list_docsets'
+        result = list_docsets(app_settings, params['arguments'] || {})
+        as_text_result(request, result)
+      when 'devdocs_search'
+        slug = params['arguments']['slug']
+        query = params['arguments']['query']
+        begin
+          entries = search_docset(app_settings, slug, query)
+          as_text_result(request, entries)
+        rescue => err
+          error(request, -32603, "Search failed: #{err.message}")
+        end
+      when 'devdocs_get_page'
+        slug = params['arguments']['slug']
+        path = params['arguments']['path']
+        begin
+          text = get_page(app_settings, slug, path)
+          respond(request, { 'content' => [{ 'type' => 'text', 'text' => text }] })
+        rescue => err
+          error(request, -32603, "Page retrieval failed: #{err.message}")
+        end
+      end
+    end
+
+    def self.list_docsets(app_settings, args)
+      offset = (args['offset'] || 0).to_i
+      limit = [(args['limit'] || 50).to_i, 500].min
+      query = args['query']&.downcase
+
+      all_docsets = app_settings.docs.values.map do |docset|
+        {
+          'slug' => docset['slug'],
+          'name' => docset['name'],
+          'version' => docset['version'],
+        }
+      end
+
+      filtered = if query
+        all_docsets.select do |docset|
+          docset['slug'].downcase.include?(query) || docset['name'].downcase.include?(query)
+        end
+      else
+        all_docsets
+      end
+
+      total_count = filtered.length
+      paginated = filtered.drop(offset).take(limit)
+
+      {
+        'docsets' => paginated,
+        'offset' => offset,
+        'limit' => limit,
+        'total' => total_count,
+        'returned' => paginated.length,
+      }
+    end
+
+    def self.validate_slug(app_settings, slug)
+      unless app_settings.docs.key?(slug)
+        raise ArgumentError, "Invalid docset slug: #{slug}"
+      end
+      slug
+    end
+
+    def self.get_page(app_settings, slug, path)
+      validate_slug(app_settings, slug)
+      db_path = File.join(app_settings.docs_path, slug, 'db.json')
+      unless File.exist?(db_path)
+        raise "Page database not available for #{slug}. Full content is served from the CDN."
+      end
+      db = JSON.parse(File.read(db_path))
+      html = db[path]
+      raise "Page not found: #{path}" unless html
+      Nokogiri::HTML::DocumentFragment.parse(html).text.squeeze(' ').strip
+    end
+
+    def self.search_docset(app_settings, slug, query)
+      validate_slug(app_settings, slug)
+      index_path = File.join(app_settings.docs_path, slug, 'index.json')
+      unless File.exist?(index_path)
+        raise "Search index not available for #{slug}. The search index is served from the CDN."
+      end
+      index = JSON.parse(File.read(index_path))
+      query_lower = query.downcase
+      index['entries'].select do |entry|
+        entry['name'].downcase.include?(query_lower) || entry['path'].downcase.include?(query_lower)
+      end
+    end
+
+    def self.as_text_result(request, data)
+      respond(request, { 'content' => [{ 'type' => 'text', 'text' => data.to_json }] })
+    end
+
+    def self.respond(request, result)
+      { 'jsonrpc' => '2.0', 'id' => request['id'], 'result' => result }
+    end
+  end
+end

test/files/docs.json

@@ -1 +1 @@
-[{"name":"CSS","slug":"css","type":"mdn","release":null,"mtime":1420139788,"db_size":3460507,"alias":null},{"name":"DOM","slug":"dom","type":"mdn","release":null,"mtime":1420139789,"db_size":11399128,"alias":null},{"name":"DOM Events","slug":"dom_events","type":"mdn","release":null,"mtime":1420139790,"db_size":889020,"alias":null},{"name":"HTML","slug":"html~5","type":"mdn","version":"5","mtime":1420139791,"db_size":1835647,"alias":null},{"name":"HTML","slug":"html~4","type":"mdn","version":"4","mtime":1420139790,"db_size":1835646,"alias":null},{"name":"HTTP","slug":"http","type":"rfc","release":null,"mtime":1420139790,"db_size":183083,"alias":null},{"name":"JavaScript","slug":"javascript","type":"mdn","release":null,"mtime":1420139791,"db_size":4125477,"alias":"js"}]
+[{"name":"CSS","slug":"css","type":"mdn","release":null,"mtime":1420139788,"db_size":3460507,"alias":null},{"name":"DOM","slug":"dom","type":"mdn","release":null,"mtime":1420139789,"db_size":11399128,"alias":null},{"name":"DOM Events","slug":"dom_events","type":"mdn","release":null,"mtime":1420139790,"db_size":889020,"alias":null},{"name":"HTML","slug":"html~5","type":"mdn","version":"5","mtime":1420139791,"db_size":1835647,"alias":null},{"name":"HTML","slug":"html~4","type":"mdn","version":"4","mtime":1420139790,"db_size":1835646,"alias":null},{"name":"HTTP","slug":"http","type":"rfc","release":null,"mtime":1420139790,"db_size":183083,"alias":null},{"name":"JavaScript","slug":"javascript","type":"mdn","release":null,"mtime":1420139791,"db_size":4125477,"alias":"js"},{"name":"MCP Fixture","slug":"mcp_fixture","type":"test","release":null,"mtime":1420139791,"db_size":1024,"alias":null}]

test/files/docs/mcp_fixture/db.json

@@ -0,0 +1 @@
+{"array/push":"<h1>Array#push</h1> <p>Appends &amp; returns the array.</p>","array/pop":"<h1>Array#pop</h1> <p>Removes the last element.</p>"}

test/files/docs/mcp_fixture/index.json

@@ -0,0 +1 @@
+{"entries":[{"name":"Array#push","path":"array/push","type":"Array"},{"name":"Array#pop","path":"array/pop","type":"Array"},{"name":"String#upcase","path":"string/upcase","type":"String"}],"types":[]}

test/mcp_test.rb

@@ -0,0 +1,184 @@
+require 'test_helper'
+require 'rack/test'
+require 'app'
+
+class McpTest < Minitest::Spec
+  include Rack::Test::Methods
+
+  def app
+    App
+  end
+
+  before do
+    current_session.env('HTTPS', 'on')
+  end
+
+  def rpc(method, params = nil, id: 1)
+    body = { jsonrpc: '2.0', id: id, method: method }
+    body[:params] = params if params
+    post '/mcp', body.to_json, 'CONTENT_TYPE' => 'application/json'
+    JSON.parse(last_response.body)
+  end
+
+  describe 'POST /mcp' do
+    it 'responds to initialize with protocol info' do
+      result = rpc('initialize')['result']
+      assert_equal '2024-11-05', result['protocolVersion']
+      assert result['capabilities'].key?('tools')
+    end
+
+    it 'lists the devdocs tools' do
+      tools = rpc('tools/list')['result']['tools']
+      names = tools.map { |t| t['name'] }
+      assert_includes names, 'devdocs_list_docsets'
+      assert_includes names, 'devdocs_search'
+      assert_includes names, 'devdocs_get_page'
+    end
+
+    it 'calls devdocs_list_docsets and returns paginated docsets in condensed format' do
+      result = rpc('tools/call', { 'name' => 'devdocs_list_docsets', 'arguments' => {} })['result']
+      response = JSON.parse(result['content'].first['text'])
+
+      assert response.key?('docsets')
+      assert response.key?('offset')
+      assert response.key?('limit')
+      assert response.key?('total')
+      assert response.key?('returned')
+
+      docsets = response['docsets']
+      assert docsets.length > 0
+      first = docsets.first
+      assert first.key?('slug')
+      assert first.key?('name')
+      assert first.key?('version')
+      refute first.key?('release_date'), 'should not include release_date'
+      refute first.key?('mtime'), 'should not include mtime'
+
+      slugs = docsets.map { |d| d['slug'] }
+      assert_includes slugs, 'css'
+      assert_includes slugs, 'html~5'
+    end
+
+    it 'paginates results with offset and limit' do
+      result = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'offset' => 0, 'limit' => 2 }
+      })['result']
+      response = JSON.parse(result['content'].first['text'])
+
+      assert_equal 0, response['offset']
+      assert_equal 2, response['limit']
+      assert_equal 2, response['returned']
+      assert response['total'] > 2
+      assert_equal 2, response['docsets'].length
+    end
+
+    it 'respects offset to skip results' do
+      first_page = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'offset' => 0, 'limit' => 2 }
+      })['result']
+      first_docsets = JSON.parse(first_page['content'].first['text'])['docsets'].map { |d| d['slug'] }
+
+      second_page = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'offset' => 2, 'limit' => 2 }
+      })['result']
+      second_docsets = JSON.parse(second_page['content'].first['text'])['docsets'].map { |d| d['slug'] }
+
+      assert first_docsets != second_docsets
+    end
+
+    it 'filters docsets by query string' do
+      result = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'query' => 'css' }
+      })['result']
+      response = JSON.parse(result['content'].first['text'])
+
+      docsets = response['docsets']
+      assert docsets.length > 0
+      assert docsets.all? { |d| d['slug'].downcase.include?('css') || d['name'].downcase.include?('css') }
+    end
+
+    it 'filters case-insensitively' do
+      result = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'query' => 'CSS' }
+      })['result']
+      response = JSON.parse(result['content'].first['text'])
+
+      docsets = response['docsets']
+      assert docsets.length > 0
+      assert docsets.any? { |d| d['slug'] == 'css' }
+    end
+
+    it 'returns empty docsets for non-matching query' do
+      result = rpc('tools/call', {
+        'name' => 'devdocs_list_docsets',
+        'arguments' => { 'query' => 'nonexistentdocthing' }
+      })['result']
+      response = JSON.parse(result['content'].first['text'])
+
+      assert_equal 0, response['returned']
+      assert_equal [], response['docsets']
+      assert response['total'] == 0
+    end
+
+    it 'calls devdocs_search and returns matching entries for a doc set' do
+      args = { 'slug' => 'mcp_fixture', 'query' => 'push' }
+      result = rpc('tools/call', { 'name' => 'devdocs_search', 'arguments' => args })['result']
+      entries = JSON.parse(result['content'].first['text'])
+      assert_equal 1, entries.length
+      assert_equal 'array/push', entries.first['path']
+    end
+
+    it 'calls devdocs_get_page and returns the entry as plain text' do
+      args = { 'slug' => 'mcp_fixture', 'path' => 'array/push' }
+      result = rpc('tools/call', { 'name' => 'devdocs_get_page', 'arguments' => args })['result']
+      text = result['content'].first['text']
+      assert_includes text, 'Array#push'
+      assert_includes text, 'Appends & returns the array.'
+      refute_includes text, '<h1>'
+    end
+
+    it 'returns error for invalid slug in search (path traversal protection)' do
+      args = { 'slug' => '../../../etc/passwd', 'query' => 'test' }
+      response = rpc('tools/call', { 'name' => 'devdocs_search', 'arguments' => args })
+      assert response.key?('error'), 'should return an error for invalid slug'
+      assert_equal(-32603, response['error']['code'])
+      assert_includes response['error']['message'], 'Invalid docset slug'
+    end
+
+    it 'returns error for invalid slug in get_page (path traversal protection)' do
+      args = { 'slug' => '..\\windows\\system32', 'path' => '/test' }
+      response = rpc('tools/call', { 'name' => 'devdocs_get_page', 'arguments' => args })
+      assert response.key?('error'), 'should return an error for invalid slug'
+      assert_equal(-32603, response['error']['code'])
+      assert_includes response['error']['message'], 'Invalid docset slug'
+    end
+
+    it 'returns error for missing search index in devdocs_search' do
+      args = { 'slug' => 'css', 'query' => 'test' }
+      response = rpc('tools/call', { 'name' => 'devdocs_search', 'arguments' => args })
+      if response.key?('error')
+        assert_equal(-32603, response['error']['code'])
+        assert_includes response['error']['message'].downcase, 'search index'
+      end
+    end
+
+    it 'returns error for missing page database in devdocs_get_page' do
+      args = { 'slug' => 'css', 'path' => '/test' }
+      response = rpc('tools/call', { 'name' => 'devdocs_get_page', 'arguments' => args })
+      if response.key?('error')
+        assert_equal(-32603, response['error']['code'])
+        assert_includes response['error']['message'].downcase, 'database'
+      end
+    end
+
+    it 'returns a JSON-RPC error for an unsupported method' do
+      response = rpc('not/a/real/method')
+      assert_equal(-32601, response['error']['code'])
+    end
+  end
+end