Search Reddit or YouTube through its official, documented API only (no scraping, no cookie import, no logged-in-human impersonation) using your own BYOK credentials, and return matching posts/videos as structured JSON. Call this whenever an agent needs recent or historical public content from one of these two platforms for research, monitoring, or compliance review; do not call it for platforms other than reddit/youtube (it returns a clear unsupported-platform error rather than guessing) or when you only need to check whether credentials work (use auth_status for that, it's cheaper and makes no search API call). Prerequisite: BYOK credentials for the target platform must already be stored locally via auditreach auth --platform <platform> -- this tool never sets up credentials itself and will fail fast with a 'No credentials found' error if they are missing. Side effects: makes a live network call to Reddit's or YouTube's official API (consuming that platform's rate limit/quota), and on success appends one tamper-evident, hash-chained entry to the local audit log recording platform, endpoint, scope, and timestamp -- this tool is not read-only and not idempotent in the sense that every successful call grows the audit log by one entry, though repeating an identical search is otherwise safe and does not mutate any remote state. If output is given, the full result set is also written to that local file path. Parameters: platform (required, 'reddit' or 'youtube'); query (search terms, required for reddit, optional for youtube); subreddit (reddit only, restrict to one subreddit, e.g. 'MachineLearning'); channel (youtube only, restrict to one channel handle, e.g. '@AnthropicAI'); since (youtube only, ISO date like '2026-06-01', only results published after this date); max_results (int, default 25, capped at 100 for reddit and 50 for youtube); before/after (reddit only, fullname pagination cursors like 't3_abc123'); output (optional local file path to also write the full JSON results to). Example calls: {"platform": "reddit", "query": "prompt injection", "subreddit": "MachineLearning", "max_results": 20}; {"platform": "youtube", "channel": "@AnthropicAI", "since": "2026-06-01", "max_results": 10}. Returns a JSON object with success (bool), exitCode (int), and on success a results array of items shaped like {id, title, url, createdAt, author, score, extra}; on failure an error string explains what went wrong (missing credentials, missing query, unsupported platform, or an upstream API error) instead of raising an exception.

Read-only check of whether BYOK credentials are already stored and still valid for a platform (equivalent to auditreach auth --platform <p> --verify --json). Call this before search to confirm credentials are configured and working, or whenever an agent needs to report connection health without touching the search API or the audit log. This tool deliberately exposes no way to set or clear credentials over MCP -- provisioning or wiping stored credentials stays a local, human-driven action via auditreach auth --platform <p> on the machine's own terminal, never something a calling agent can trigger remotely. Behaviorally, this is read-only: it makes a lightweight verification call against the platform's API to confirm the stored credential still works, but it never writes to the audit log and never mutates local credential storage. It is idempotent and safe to call repeatedly. Parameters: platform (required, 'reddit' or 'youtube'; any other value returns a structured unsupported-platform error rather than failing silently). Example calls: {"platform": "reddit"}; {"platform": "youtube"}. Returns a JSON object with success (bool) and exitCode (int); on success the payload confirms credentials are present and valid, on failure error explains what's missing (e.g. no credentials stored, or the stored credential was rejected by the platform).

Verify that the local hash-chained audit log has not been tampered with, by walking every entry and confirming each one's hash correctly chains to the previous entry (equivalent to auditreach verify-log --json). Call this to prove compliance/audit integrity before sharing the log with a third party, after any manual edit to the log file, or periodically as a trust check -- it does not need to run before every search. This tool is fully read-only: it opens and reads the log file but never writes to it, is idempotent, and produces the same verdict on repeated calls against an unchanged file. Prerequisite: an audit log file must already exist (one is created automatically the first time search succeeds); pointing this at a path with no log file returns a structured failure rather than raising. Parameters: path (optional string; defaults to ./auditreach.log.jsonl in the current working directory if omitted). Example calls: {} (verify the default log); {"path": "./auditreach.log.jsonl"}. Returns a JSON object with success (bool, true only if the entire chain verifies intact) and exitCode (int); on failure error describes the problem, such as a broken hash link at a specific entry index or a missing/unreadable log file.