ketch

Backends ​

ketch has three search surfaces, each with its own backends: web search (ketch search), code search (ketch code), and library docs (ketch docs).

Web Search Backends ​

ketch supports six web-search backends. Set the default with ketch config set backend <name>.

Brave (default) ​

Brave Search offers a free API tier — no scraping, proper JSON API, reliable.

Setup:

  1. Get a free API key at brave.com/search/api
  2. Set it: ketch config set brave_api_key <your-key>

Free tier limits: 2,000 queries/month, 1 query/second.

DuckDuckGo ​

Zero-config HTML scraping of DuckDuckGo's search results. No API key needed.

Setup: None — works out of the box.

Limitations: DuckDuckGo aggressively rate-limits automated requests. You may see ddg rate limited after retries errors under heavy use. ketch retries up to 3 times with 500ms backoff.

SearXNG ​

Self-hosted metasearch engine with a JSON API. The most reliable option for heavy use.

Setup:

  1. Run a SearXNG instance (Docker is easiest):
sh
docker run -d -p 8081:8080 searxng/searxng

  1. Enable JSON format in SearXNG settings (required for the API).

  2. Point ketch to it:

sh
ketch config set backend searxng
ketch config set searxng_url http://localhost:8081

Recommended for: operators running agents that search frequently, or anyone who wants full control over their search infrastructure.

Exa ​

AI-oriented web search via Exa's hosted MCP endpoint. It works without configuration by default, with an optional Exa API key for authenticated usage.

Setup: None for hosted MCP. Optional key:

sh
ketch config set exa_api_key <your-key>
ketch config set backend exa

Recommended for: agent workflows that benefit from Exa's clean result snippets and content-oriented search output.

Firecrawl ​

Web search via the Firecrawl v2 search API — proper JSON API, no scraping. Requires an API key.

Setup:

  1. Get an API key at firecrawl.dev
  2. Set it: ketch config set firecrawl_api_key <your-key>
  3. Make it the default: ketch config set backend firecrawl

Recommended for: operators already using Firecrawl for scraping who want a single provider for both search and page extraction. Pair with --scrape to fetch full content per result.

Keenable ​

Web search built for AI agents, backed by the Keenable index. Keyless by default — it works with no account or key against the public endpoint (rate-limited); an optional API key lifts the hourly cap.

Setup: None. Optional key to lift the rate limit:

sh
ketch config set keenable_api_key <your-key>
ketch config set backend keenable

Create a key at keenable.ai/console.

Recommended for: agent workflows that want a zero-config, agent-oriented search backend without provisioning a provider key up front.

Code Search Backends ​

ketch code searches real source code across open-source repositories. Set the default with ketch config set code_backend <name>.

Grep (default) ​

The Grep MCP server (mcp.grep.app) — literal or regex search over 1M+ public GitHub repos. Zero config, no token.

Setup: None — works out of the box.

Use --regex to interpret the query as a regular expression.

Sourcegraph ​

Grep-style search across ~1M OSS repos with exact line matches over an SSE stream. Results are filtered to non-archived, non-fork repos by default. Supports --regex.

Setup: None for the public instance. For a self-hosted instance: ketch config set sourcegraph_url <url>.

GitHub ​

GitHub Code Search (REST /search/code) with a batched GraphQL call for star counts.

Setup: A token is required. Resolution chain: ketch config set github_token <tok> → $GITHUB_TOKEN → $GH_TOKEN → gh auth token (if the gh CLI is installed).

Limits: 30 requests/minute. Token must have repo scope.

Docs Backends ​

ketch docs fetches library documentation. Set the default with ketch config set docs_backend <name>.

Context7 (default) ​

Curated, version-aware documentation snippets.

Setup: Free key: ketch config set context7_api_key <key>.

Local ​

A planned FTS5 SQLite backend for offline/private docs. Not yet implemented.