majiayu000 Keenable Search Skill
Web search and page fetch via Keenable AI — purpose-built for AI agents. Two scripts cover all use cases: search the web, fetch/extract page contents.
Prerequisite: KEENABLE_API_KEY environment variable. Get key at https://app.keenable.ai
When to Use Keenable
- Default web search — 100K free requests/month, generous rate limits
- Point-in-time search — query the index as it existed at a past date (
--query-time) - Prompt-driven extraction — fetch a URL and have the API extract exactly what you need (
--prompt) - High-volume research — 100x the free tier of Exa (100K vs 1K/month)
Token-Efficient Research
Search cheaply, pick the best results, extract selectively — reduces context by filtering before reasoning.
# Step 1: Search with short snippets to scan titles/descriptions
python3 ~/.claude/skills/keenable/scripts/keenable_search.py "query" --count 10 --snippet-length 200
# Step 2: Pick the 2-3 most relevant URLs from results
# Step 3: Extract only what you need from those URLs
python3 ~/.claude/skills/keenable/scripts/keenable_fetch.py URL --prompt "Extract the key findings"
Available Scripts
1. keenable_search.py — Web Search
python3 ~/.claude/skills/keenable/scripts/keenable_search.py "query" [options]
| Quick Example | Purpose |
|---|---|
... keenable_search.py "AI agents 2026" |
Basic search |
... keenable_search.py "Claude Code" --site docs.anthropic.com |
Site-restricted |
... keenable_search.py "transformer paper" --count 20 |
More results |
... keenable_search.py "election results" --published-after 7d |
Recent only |
... keenable_search.py "AI safety" --query-time 2026-01-01 |
Historical index |
... keenable_search.py "LLM benchmarks" --snippet-length 5000 |
Longer snippets |
Key options:
--site DOMAIN— restrict to a specific domain--count N— number of results (1-50, default 10)--snippet-length N— snippet max chars (180-10000)--published-after DATE— filter by publish date (absolute or relative:7d,30min,2h)--published-before DATE— upper bound on publish date--query-time DATETIME— search the index as it stood at this point in time--json— raw JSON output (default: formatted table)
2. keenable_fetch.py — Page Fetch & Extraction
python3 ~/.claude/skills/keenable/scripts/keenable_fetch.py URL [options]
| Quick Example | Purpose |
|---|---|
... keenable_fetch.py "https://example.com" |
Fetch as markdown |
... keenable_fetch.py URL --live |
Force live fetch (not cached) |
... keenable_fetch.py URL --prompt "List all pricing tiers" |
Extract specific info |
... keenable_fetch.py URL --max-chars 5000 |
Bounded output |
... keenable_fetch.py URL --live --prompt "What is the main argument?" |
Live + extract |
Key options:
--live— fetch live page instead of indexed/cached copy--prompt TEXT— LLM-driven extraction (API reads page, returns only what you asked for)--max-chars N— truncate output to N characters--json— raw JSON response
3. keenable_setup.py — Setup & Validation
python3 ~/.claude/skills/keenable/scripts/keenable_setup.py --validate # Test API key
python3 ~/.claude/skills/keenable/scripts/keenable_setup.py --setup-mcp # Add MCP server
Script Selection Guide
| Task | Best Approach |
|---|---|
| General web search | keenable_search.py "query" |
| Search specific site | keenable_search.py "query" --site domain.com |
| Recent news/articles | keenable_search.py "topic" --published-after 7d |
| Historical search | keenable_search.py "query" --query-time 2025-06-01 |
| Read a known URL | keenable_fetch.py URL |
| Extract specific data from URL | keenable_fetch.py URL --prompt "extract X" |
| Fresh content (bypass cache) | keenable_fetch.py URL --live |
| Bounded page read | keenable_fetch.py URL --max-chars 3000 |
Keenable vs Exa vs Firecrawl
| Need | Best Tool | Why |
|---|---|---|
| General web search | Keenable | 100K free/mo, fast, reliable |
| Point-in-time search | Keenable | Only tool that supports historical index queries |
| Extract specific data from URL | Keenable --prompt |
API-side LLM extraction, no post-processing |
| Semantic/neural search | Exa | AI-powered relevance ranking |
| Find similar pages | Exa exa_similar.py |
Semantic similarity |
| Deep site crawling | Firecrawl | Link following, JS rendering |
| Search + scrape combined | Firecrawl search --scrape |
One operation |
| High-volume usage | Keenable | 100K vs Exa 1K vs Firecrawl 500 |
Common Workflows
Search then Read
# Find relevant pages
python3 ~/.claude/skills/keenable/scripts/keenable_search.py "React Server Components" --count 5
# Read the most relevant result
python3 ~/.claude/skills/keenable/scripts/keenable_fetch.py "https://react.dev/blog/..." --max-chars 8000
Extract Structured Data
python3 ~/.claude/skills/keenable/scripts/keenable_fetch.py "https://example.com/pricing" \
--prompt "List all pricing tiers with name, price, and features as a markdown table"
Research with Date Bounds
python3 ~/.claude/skills/keenable/scripts/keenable_search.py "AI regulation" \
--published-after 2026-01-01 --published-before 2026-06-01 --count 20
MCP Alternative
For native tool access without scripts, add Keenable as an MCP server:
claude mcp add keenable \
--transport http https://api.keenable.ai/mcp \
--scope user \
--header "X-API-Key: $KEENABLE_API_KEY"
Troubleshooting
| Problem | Fix |
|---|---|
KEENABLE_API_KEY not set |
Export in ~/.config/env/secrets.env or shell profile |
401 Unauthorized |
Check key starts with keen_ and is valid |
429 Rate Limited |
Authenticated: per-org limits. Unauthenticated: 1K/hr shared |
| Stale results | Use --live flag on fetch, or remove --query-time from search |
| Truncated content | Increase --max-chars or --snippet-length |
| Error fetching URL (404/422) | URL may not be indexed — add --live to fetch any URL |
Reference
| File | Contents |
|---|---|
references/api-reference.md |
Full API parameter reference for both endpoints |
references/selection-guide.md |
Detailed comparison with Exa and Firecrawl |