1package tools
2
3import (
4 "context"
5 "encoding/json"
6 "fmt"
7 "io"
8 "net/http"
9 "strings"
10 "time"
11
12 md "github.com/JohannesKaufmann/html-to-markdown"
13 "github.com/PuerkitoBio/goquery"
14 "github.com/charmbracelet/crush/internal/config"
15 "github.com/charmbracelet/crush/internal/permission"
16)
17
18type FetchParams struct {
19 URL string `json:"url"`
20 Format string `json:"format"`
21 Timeout int `json:"timeout,omitempty"`
22}
23
24type FetchPermissionsParams struct {
25 URL string `json:"url"`
26 Format string `json:"format"`
27 Timeout int `json:"timeout,omitempty"`
28}
29
30type fetchTool struct {
31 client *http.Client
32 permissions permission.Service
33}
34
35const (
36 FetchToolName = "fetch"
37 fetchToolDescription = `Fetches content from a URL and returns it in the specified format.
38
39WHEN TO USE THIS TOOL:
40- Use when you need to download content from a URL
41- Helpful for retrieving documentation, API responses, or web content
42- Useful for getting external information to assist with tasks
43
44HOW TO USE:
45- Provide the URL to fetch content from
46- Specify the desired output format (text, markdown, or html)
47- Optionally set a timeout for the request
48
49FEATURES:
50- Supports three output formats: text, markdown, and html
51- Automatically handles HTTP redirects
52- Sets reasonable timeouts to prevent hanging
53- Validates input parameters before making requests
54
55LIMITATIONS:
56- Maximum response size is 5MB
57- Only supports HTTP and HTTPS protocols
58- Cannot handle authentication or cookies
59- Some websites may block automated requests
60
61TIPS:
62- Use text format for plain text content or simple API responses
63- Use markdown format for content that should be rendered with formatting
64- Use html format when you need the raw HTML structure
65- Set appropriate timeouts for potentially slow websites`
66)
67
68func NewFetchTool(permissions permission.Service) BaseTool {
69 return &fetchTool{
70 client: &http.Client{
71 Timeout: 30 * time.Second,
72 Transport: &http.Transport{
73 MaxIdleConns: 100,
74 MaxIdleConnsPerHost: 10,
75 IdleConnTimeout: 90 * time.Second,
76 },
77 },
78 permissions: permissions,
79 }
80}
81
82func (t *fetchTool) Info() ToolInfo {
83 return ToolInfo{
84 Name: FetchToolName,
85 Description: fetchToolDescription,
86 Parameters: map[string]any{
87 "url": map[string]any{
88 "type": "string",
89 "description": "The URL to fetch content from",
90 },
91 "format": map[string]any{
92 "type": "string",
93 "description": "The format to return the content in (text, markdown, or html)",
94 "enum": []string{"text", "markdown", "html"},
95 },
96 "timeout": map[string]any{
97 "type": "number",
98 "description": "Optional timeout in seconds (max 120)",
99 },
100 },
101 Required: []string{"url", "format"},
102 }
103}
104
105func (t *fetchTool) Run(ctx context.Context, call ToolCall) (ToolResponse, error) {
106 var params FetchParams
107 if err := json.Unmarshal([]byte(call.Input), ¶ms); err != nil {
108 return NewTextErrorResponse("Failed to parse fetch parameters: " + err.Error()), nil
109 }
110
111 if params.URL == "" {
112 return NewTextErrorResponse("URL parameter is required"), nil
113 }
114
115 format := strings.ToLower(params.Format)
116 if format != "text" && format != "markdown" && format != "html" {
117 return NewTextErrorResponse("Format must be one of: text, markdown, html"), nil
118 }
119
120 if !strings.HasPrefix(params.URL, "http://") && !strings.HasPrefix(params.URL, "https://") {
121 return NewTextErrorResponse("URL must start with http:// or https://"), nil
122 }
123
124 sessionID, messageID := GetContextValues(ctx)
125 if sessionID == "" || messageID == "" {
126 return ToolResponse{}, fmt.Errorf("session ID and message ID are required for creating a new file")
127 }
128
129 p := t.permissions.Request(
130 permission.CreatePermissionRequest{
131 SessionID: sessionID,
132 Path: config.WorkingDirectory(),
133 ToolName: FetchToolName,
134 Action: "fetch",
135 Description: fmt.Sprintf("Fetch content from URL: %s", params.URL),
136 Params: FetchPermissionsParams(params),
137 },
138 )
139
140 if !p {
141 return ToolResponse{}, permission.ErrorPermissionDenied
142 }
143
144 // Handle timeout with context
145 requestCtx := ctx
146 if params.Timeout > 0 {
147 maxTimeout := 120 // 2 minutes
148 if params.Timeout > maxTimeout {
149 params.Timeout = maxTimeout
150 }
151 var cancel context.CancelFunc
152 requestCtx, cancel = context.WithTimeout(ctx, time.Duration(params.Timeout)*time.Second)
153 defer cancel()
154 }
155
156 req, err := http.NewRequestWithContext(requestCtx, "GET", params.URL, nil)
157 if err != nil {
158 return ToolResponse{}, fmt.Errorf("failed to create request: %w", err)
159 }
160
161 req.Header.Set("User-Agent", "crush/1.0")
162
163 resp, err := t.client.Do(req)
164 if err != nil {
165 return ToolResponse{}, fmt.Errorf("failed to fetch URL: %w", err)
166 }
167 defer resp.Body.Close()
168
169 if resp.StatusCode != http.StatusOK {
170 return NewTextErrorResponse(fmt.Sprintf("Request failed with status code: %d", resp.StatusCode)), nil
171 }
172
173 maxSize := int64(5 * 1024 * 1024) // 5MB
174 body, err := io.ReadAll(io.LimitReader(resp.Body, maxSize))
175 if err != nil {
176 return NewTextErrorResponse("Failed to read response body: " + err.Error()), nil
177 }
178
179 content := string(body)
180 contentType := resp.Header.Get("Content-Type")
181
182 switch format {
183 case "text":
184 if strings.Contains(contentType, "text/html") {
185 text, err := extractTextFromHTML(content)
186 if err != nil {
187 return NewTextErrorResponse("Failed to extract text from HTML: " + err.Error()), nil
188 }
189 return NewTextResponse(text), nil
190 }
191 return NewTextResponse(content), nil
192
193 case "markdown":
194 if strings.Contains(contentType, "text/html") {
195 markdown, err := convertHTMLToMarkdown(content)
196 if err != nil {
197 return NewTextErrorResponse("Failed to convert HTML to Markdown: " + err.Error()), nil
198 }
199 return NewTextResponse(markdown), nil
200 }
201
202 return NewTextResponse("```\n" + content + "\n```"), nil
203
204 case "html":
205 return NewTextResponse(content), nil
206
207 default:
208 return NewTextResponse(content), nil
209 }
210}
211
212func extractTextFromHTML(html string) (string, error) {
213 doc, err := goquery.NewDocumentFromReader(strings.NewReader(html))
214 if err != nil {
215 return "", err
216 }
217
218 text := doc.Text()
219 text = strings.Join(strings.Fields(text), " ")
220
221 return text, nil
222}
223
224func convertHTMLToMarkdown(html string) (string, error) {
225 converter := md.NewConverter("", true, nil)
226
227 markdown, err := converter.ConvertString(html)
228 if err != nil {
229 return "", err
230 }
231
232 return markdown, nil
233}