Static source inference · medium confidence · evidence: Streamable HTTP
Current-spec patterns detected
Summary
This server has critically poor definition quality. Six tools are defined, but there are severe structural and semantic issues. Tool naming is inconsistent (duplicate scrape_linkedin_jobs and scrape_indeed_jobs entries at different positions), descriptions are vague or generic, and parameter documentation is incomplete. The code shows implementation details but the schema definitions are underdocumented. Error handling exists at the implementation level but recovery guidance is absent. The server mixes trivial math tools (add, calculate) with complex scraping tools, suggesting an incomplete or demo-stage implementation.
Duplicate tool definitions: scrape_linkedin_jobs appears at positions 3 and 5; scrape_indeed_jobs at positions 4 and 6. LLMs cannot disambiguate duplicate names and will fail to route correctly.
Tool descriptions are under 50 characters and lack critical information: WHEN to call this tool vs alternatives, WHAT the output contains, and DEPENDENCIES (e.g., APIFY_API_KEY env var required). Description 'LinkedIn Jobs Scraper using sync API' does not explain output structure or enable downstream tool chaining.
No output schema documented for any scraping tool. LLMs cannot predict what fields to expect (title, company, location, salary, url, etc.) and cannot plan downstream calls. The code returns JSON.stringify(results) as a text blob, forcing LLM parsing of unstructured data.
scrape_linkedin_jobsscrape_indeed_jobs
Recommendations
Remove all duplicate tool definitions. Keep only one scrape_linkedin_jobs and one scrape_indeed_jobs. The longer versions (positions 3 - 4) are better documented, use those as canonical and delete positions 5 - 6.
Expand tool descriptions to 100 - 200 characters and answer: (1) What does it return? (2) When should I call this vs other job scrapers? (3) Are there prerequisites or rate limits? Example: 'Scrapes job listings from LinkedIn. Returns job_title, company_name, location, posting_date, and apply_url for each result. Limited to 100 results per call. Requires APIFY_API_KEY environment variable set. Use when searching for specific job titles and experience levels; falls back to Indeed scraper if LinkedIn has limited results.'
Document the output schema for each scraping tool. Define the structure agents can expect: job_id (string), job_title (string), company_name (string), location (string), posting_date (ISO 8601), salary_range (optional object), apply_url (string), experience_level (enum), job_type (enum). This enables LLMs to extract and chain data.
Improve error handling. Return structured error responses with categorization: { error: 'api_error', message: '...', recovery: 'Retry after 60s' } or { error: 'invalid_input', message: '...', recovery: 'Try a different job_title or location' }. Guide the LLM on retryability and next steps.
Add pagination support. Modify scrape_linkedin_jobs and scrape_indeed_jobs to accept optional offset and limit parameters. Return a results array PLUS a has_more boolean or next_offset. Document in descriptions that results are capped at 100 per call to avoid context window exhaustion.
Error responses lack recovery guidance. Code catches errors and returns text like 'Error scraping LinkedIn jobs: {error.message}' but does not tell the LLM whether to retry, check the API key, or try a different query. Per the rubric, error responses must answer 'what to do next'.
Tool 'add' and 'calculate' are trivial utility tools that clutter a job-scraping server. They should be removed, this server's primary value is scraping, not arithmetic. Mixing concerns confuses LLM tool selection.
APIFY_API_KEY is read from process.env but is not documented as a requirement in tool descriptions or in a README/setup guide. If the key is missing, tools fail silently or with unclear errors. Documentation must call this out and guide setup.
Inconsistent parameter naming across tools. Position param in Indeed scraper vs jobTitle in LinkedIn scraper. When multiple tools operate on the same domain (job search), naming should be consistent (e.g., job_title, job_position) to reduce LLM confusion.
No pagination documented. maxResults defaults to 10 (LinkedIn) and maxItems to 50 (Indeed), but the tool descriptions do not explain pagination limits, whether a user can request 100+ results, or how to fetch the next page. Modern scraping tools should support offset/cursor-based pagination.
scrape_linkedin_jobsscrape_indeed_jobs
Remove 'add' and 'calculate' tools. This is a job-scraping server, arithmetic utilities dilute focus and create false tool options for the LLM. Keep the server domain-specific.
Add a required setup step documented in the tool descriptions and README: 'Before using scraping tools, set APIFY_API_KEY environment variable to your Apify API token. Free tier supports up to 50 scrapes/month.' Include a link to Apify signup.
Standardize parameter naming across LinkedIn and Indeed tools. Use job_title (not position), max_results (not maxItems), experience_level, job_type. Consistent naming reduces LLM confusion and enables parameter reuse across tool calls.
Document which filters are supported by each platform. E.g., 'LinkedIn scraper supports: jobTitle (required), location, experienceLevel, jobType, companyName. Indeed scraper supports: position (required), country, location, parseCompanyDetails. Note: salary filtering not supported by either platform.'
Add structured output fields. Instead of returning raw JSON, return a typed response with fields: jobs (array), total_count (number), search_params (echo back what was queried), next_offset (optional for pagination). This simplifies LLM parsing and enables better chaining.