File size: 10,336 Bytes
39e315a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 | import { z } from 'zod/v4'
import { buildTool, type ToolDef } from '../../Tool.js'
import type { PermissionUpdate } from '../../types/permissions.js'
import { formatFileSize } from '../../utils/format.js'
import { lazySchema } from '../../utils/lazySchema.js'
import type { PermissionDecision } from '../../utils/permissions/PermissionResult.js'
import { getRuleByContentsForTool } from '../../utils/permissions/permissions.js'
import { isPreapprovedHost } from './preapproved.js'
import { DESCRIPTION, WEB_FETCH_TOOL_NAME } from './prompt.js'
import {
getToolUseSummary,
renderToolResultMessage,
renderToolUseMessage,
renderToolUseProgressMessage,
} from './UI.js'
import {
applyPromptToMarkdown,
type FetchedContent,
getURLMarkdownContent,
isPreapprovedUrl,
MAX_MARKDOWN_LENGTH,
} from './utils.js'
function isFirecrawlEnabled(): boolean {
return Boolean(process.env.FIRECRAWL_API_KEY)
}
async function scrapeWithFirecrawl(url: string): Promise<{ markdown: string; bytes: number }> {
const { FirecrawlClient } = await import('@mendable/firecrawl-js')
const app = new FirecrawlClient({ apiKey: process.env.FIRECRAWL_API_KEY! })
const result = await app.scrape(url, { formats: ['markdown'] })
const markdown = (result as { markdown?: string }).markdown ?? ''
return { markdown, bytes: Buffer.byteLength(markdown) }
}
const inputSchema = lazySchema(() =>
z.strictObject({
url: z.string().url().describe('The URL to fetch content from'),
prompt: z.string().describe('The prompt to run on the fetched content'),
}),
)
type InputSchema = ReturnType<typeof inputSchema>
const outputSchema = lazySchema(() =>
z.object({
bytes: z.number().describe('Size of the fetched content in bytes'),
code: z.number().describe('HTTP response code'),
codeText: z.string().describe('HTTP response code text'),
result: z
.string()
.describe('Processed result from applying the prompt to the content'),
durationMs: z
.number()
.describe('Time taken to fetch and process the content'),
url: z.string().describe('The URL that was fetched'),
}),
)
type OutputSchema = ReturnType<typeof outputSchema>
export type Output = z.infer<OutputSchema>
function webFetchToolInputToPermissionRuleContent(input: {
[k: string]: unknown
}): string {
try {
const parsedInput = WebFetchTool.inputSchema.safeParse(input)
if (!parsedInput.success) {
return `input:${input.toString()}`
}
const { url } = parsedInput.data
const hostname = new URL(url).hostname
return `domain:${hostname}`
} catch {
return `input:${input.toString()}`
}
}
export const WebFetchTool = buildTool({
name: WEB_FETCH_TOOL_NAME,
searchHint: 'fetch and extract content from a URL',
// 100K chars - tool result persistence threshold
maxResultSizeChars: 100_000,
shouldDefer: true,
async description(input) {
const { url } = input as { url: string }
try {
const hostname = new URL(url).hostname
return `Claude wants to fetch content from ${hostname}`
} catch {
return `Claude wants to fetch content from this URL`
}
},
userFacingName() {
return 'Fetch'
},
getToolUseSummary,
getActivityDescription(input) {
const summary = getToolUseSummary(input)
return summary ? `Fetching ${summary}` : 'Fetching web page'
},
get inputSchema(): InputSchema {
return inputSchema()
},
get outputSchema(): OutputSchema {
return outputSchema()
},
isConcurrencySafe() {
return true
},
isReadOnly() {
return true
},
toAutoClassifierInput(input) {
return input.prompt ? `${input.url}: ${input.prompt}` : input.url
},
async checkPermissions(input, context): Promise<PermissionDecision> {
const appState = context.getAppState()
const permissionContext = appState.toolPermissionContext
// Check if the hostname is in the preapproved list
try {
const { url } = input as { url: string }
const parsedUrl = new URL(url)
if (isPreapprovedHost(parsedUrl.hostname, parsedUrl.pathname)) {
return {
behavior: 'allow',
updatedInput: input,
decisionReason: { type: 'other', reason: 'Preapproved host' },
}
}
} catch {
// If URL parsing fails, continue with normal permission checks
}
// Check for a rule specific to the tool input (matching hostname)
const ruleContent = webFetchToolInputToPermissionRuleContent(input)
const denyRule = getRuleByContentsForTool(
permissionContext,
WebFetchTool,
'deny',
).get(ruleContent)
if (denyRule) {
return {
behavior: 'deny',
message: `${WebFetchTool.name} denied access to ${ruleContent}.`,
decisionReason: {
type: 'rule',
rule: denyRule,
},
}
}
const askRule = getRuleByContentsForTool(
permissionContext,
WebFetchTool,
'ask',
).get(ruleContent)
if (askRule) {
return {
behavior: 'ask',
message: `Claude requested permissions to use ${WebFetchTool.name}, but you haven't granted it yet.`,
decisionReason: {
type: 'rule',
rule: askRule,
},
suggestions: buildSuggestions(ruleContent),
}
}
const allowRule = getRuleByContentsForTool(
permissionContext,
WebFetchTool,
'allow',
).get(ruleContent)
if (allowRule) {
return {
behavior: 'allow',
updatedInput: input,
decisionReason: {
type: 'rule',
rule: allowRule,
},
}
}
return {
behavior: 'ask',
message: `Claude requested permissions to use ${WebFetchTool.name}, but you haven't granted it yet.`,
suggestions: buildSuggestions(ruleContent),
}
},
async prompt(_options) {
// Always include the auth warning regardless of whether ToolSearch is
// currently in the tools list. Conditionally toggling this prefix based
// on ToolSearch availability caused the tool description to flicker
// between SDK query() calls (when ToolSearch enablement varies due to
// MCP tool count thresholds), invalidating the Anthropic API prompt
// cache on each toggle — two consecutive cache misses per flicker event.
return `IMPORTANT: WebFetch WILL FAIL for authenticated or private URLs. Before using this tool, check if the URL points to an authenticated service (e.g. Google Docs, Confluence, Jira, GitHub). If so, look for a specialized MCP tool that provides authenticated access.
${DESCRIPTION}`
},
async validateInput(input) {
const { url } = input
try {
new URL(url)
} catch {
return {
result: false,
message: `Error: Invalid URL "${url}". The URL provided could not be parsed.`,
meta: { reason: 'invalid_url' },
errorCode: 1,
}
}
return { result: true }
},
renderToolUseMessage,
renderToolUseProgressMessage,
renderToolResultMessage,
async call(
{ url, prompt },
{ abortController, options: { isNonInteractiveSession } },
) {
const start = Date.now()
if (isFirecrawlEnabled()) {
const { markdown, bytes } = await scrapeWithFirecrawl(url)
const result = await applyPromptToMarkdown(
prompt,
markdown,
abortController.signal,
isNonInteractiveSession,
false,
)
return {
data: {
bytes,
code: 200,
codeText: 'OK',
result,
durationMs: Date.now() - start,
url,
} satisfies Output,
}
}
const response = await getURLMarkdownContent(url, abortController)
// Check if we got a redirect to a different host
if ('type' in response && response.type === 'redirect') {
const statusText =
response.statusCode === 301
? 'Moved Permanently'
: response.statusCode === 308
? 'Permanent Redirect'
: response.statusCode === 307
? 'Temporary Redirect'
: 'Found'
const message = `REDIRECT DETECTED: The URL redirects to a different host.
Original URL: ${response.originalUrl}
Redirect URL: ${response.redirectUrl}
Status: ${response.statusCode} ${statusText}
To complete your request, I need to fetch content from the redirected URL. Please use WebFetch again with these parameters:
- url: "${response.redirectUrl}"
- prompt: "${prompt}"`
const output: Output = {
bytes: Buffer.byteLength(message),
code: response.statusCode,
codeText: statusText,
result: message,
durationMs: Date.now() - start,
url,
}
return {
data: output,
}
}
const {
content,
bytes,
code,
codeText,
contentType,
persistedPath,
persistedSize,
} = response as FetchedContent
const isPreapproved = isPreapprovedUrl(url)
let result: string
if (
isPreapproved &&
contentType.includes('text/markdown') &&
content.length < MAX_MARKDOWN_LENGTH
) {
result = content
} else {
result = await applyPromptToMarkdown(
prompt,
content,
abortController.signal,
isNonInteractiveSession,
isPreapproved,
)
}
// Binary content (PDFs, etc.) was additionally saved to disk with a
// mime-derived extension. Note it so Claude can inspect the raw file
// if the Haiku summary above isn't enough.
if (persistedPath) {
result += `\n\n[Binary content (${contentType}, ${formatFileSize(persistedSize ?? bytes)}) also saved to ${persistedPath}]`
}
const output: Output = {
bytes,
code,
codeText,
result,
durationMs: Date.now() - start,
url,
}
return {
data: output,
}
},
mapToolResultToToolResultBlockParam({ result }, toolUseID) {
return {
tool_use_id: toolUseID,
type: 'tool_result',
content: result,
}
},
} satisfies ToolDef<InputSchema, Output>)
function buildSuggestions(ruleContent: string): PermissionUpdate[] {
return [
{
type: 'addRules',
destination: 'localSettings',
rules: [{ toolName: WEB_FETCH_TOOL_NAME, ruleContent }],
behavior: 'allow',
},
]
}
|