PulseLink — keyless open-web & social research source (MCP Server + A2A Server). Search, fetch, list, and transcribe across YouTube, Reddit, X, Hacker News, the web, RSS/news, GitHub and more; pulse_status reports per-source health.
PulseLink defines 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (pulse_search, pulse_fetch, pulse_list, pulse_transcribe, pulse_status, pulse_ingest, x_search, browse_x_post). However, descriptions are inconsistent in depth and actionability. Most tool descriptions are 50-150 chars, which is acceptable, but lack guidance on when to use each tool vs. alternatives (e.g., pulse_search vs pulse_list vs pulse_ingest). Parameter descriptions are present but generic (e.g., 'Source to search' without enumerating valid sources). Output schemas are not documented in the provided code. Error handling and recovery guidance are absent. The x_search tool has good parameter validation (mutually exclusive handles, date format hints) but pulse_* tools lack similar constraints. No tool annotations (readOnlyHint/destructiveHint) are visible, despite pulse_ingest being a WRITE operation.
Browse and extract detailed information from a specific X post using xAI's Grok index.
Fetch one item (full text/body/transcript). CONCEPT:PK-OS.governance.search-fetch-list-transcribe
Search/list a source and natively ingest the results into epistemic-graph. Lists via the real PulseLink client (pulse_list when channel is set, else pulse_search) and pushes the documents as typed :Document nodes linked to their :PulseSource and :Person author. Native-ingest failures propagate to the caller. CONCEPT:AU-KG.ingest.enterprise-source-extractor
List items from a source channel/feed. CONCEPT:PK-OS.governance.search-fetch-list-transcribe
Search a source and return normalized documents. CONCEPT:PK-OS.governance.search-fetch-list-transcribe
Per-source backend + credential health (the doctor). CONCEPT:PK-OS.governance.search-fetch-list-transcribe
Output schemas not documented. Tool descriptions state what is returned (e.g., 'normalized documents', 'full text') but do not specify field names, types, or structure. LLMs cannot plan downstream tool calls or extract required data without knowing response shape.
Parameter descriptions lack actionable constraints. 'Source to search. One of: youtube, reddit, x, hackernews, web, rss, github, etc.' is a free-form list, not an enum. LLMs may hallucinate invalid sources. Should declare source as an enum with exactly valid values.
No tool annotations visible. pulse_ingest is marked WRITE but has no destructiveHint annotation. pulse_search, pulse_fetch, pulse_list, pulse_transcribe, pulse_status, x_search, browse_x_post are READ_ONLY but lack readOnlyHint. Annotations help LLMs reason about safety and retry logic.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | <=2025-11-25 | v2 |
Transcribe video/audio to text. CONCEPT:PK-OS.governance.audio-video-sources-transcript
Search X (formerly Twitter) posts, profiles, and threads using xAI's native Grok index. This leverages the live X search tool via xAI, returning detailed summaries and citations. Exactly one of allowed_x_handles or excluded_x_handles may be specified, but not both.
No error handling or recovery guidance. Tools lack descriptions of failure modes (e.g., 'Source not available', 'Transcription failed', 'Ingest conflict'). LLMs receive no guidance on what to do next when a call fails.
Tool selection ambiguity. pulse_search, pulse_list, and pulse_ingest all retrieve content from sources. Descriptions do not clarify when to use each: pulse_search queries by keyword, pulse_list browses a channel, pulse_ingest does both and writes to graph. LLMs may pick the wrong tool.