A self-hosted, open-source RSS reader with a graph-native data model, REST API, and MCP server
Reed demonstrates solid tool design with 21 well-named, verb-prefixed tools covering feed management, item operations, and guided workflows. All tools have descriptions (avg ~120 chars, within baseline 194 chars). Input schemas are present and typed for all tools. However, output schemas are not documented in the visible code, LLMs cannot predict return structures. Error handling is minimal; most tools lack recovery guidance. Three workflow tools (deep_reading_session, weekly_digest, research_thread) are prompts, not traditional tools, creating composition ambiguity. Parameter descriptions are adequate but lack constraint details (e.g., 'limit' lacks explicit bounds in descriptions, though code validates 1-200). No tool annotations (readOnlyHint, destructiveHint) despite clear risk classifications.
Creates or replaces the note on an item. content: null deletes the note.
Guided workflow prompt for deep reading: select an unread item, read it fully, and optionally annotate or star it.
Export all subscribed feeds as OPML (Outline Processor Markup Language).
Returns items similar to a given item, via SIMILAR_TO edges.
Return Reed's effective global configuration.
Return a single feed with full metadata.
Output schemas not documented. LLMs cannot predict return structures for any tool. Visible code shows input schemas but no @mcp.tool return type hints or docstring schema documentation.
Three workflow tools (deep_reading_session, weekly_digest, research_thread) are prompts masquerading as tools with empty input schemas. Composition is ambiguous, LLMs cannot chain these with other tools or pass parameters.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk classifications exist in metadata but are not exposed to LLMs via MCP tool annotations, forcing LLMs to infer safety from descriptions alone.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Return a single item with full content. Use this to read an article.
Return items matching filters. The primary reading tool. With no filters, returns the most recent items across all feeds.
Return a single topic with its items.
Import feeds from an OPML document. Returns the list of created feeds.
Return all subscribed feeds with metadata and unread counts.
Return all tags in the system with usage counts.
Return all topics (auto-discovered clusters) with item counts.
Mark one or more items as read. item_ids, feed_id, and before are all optional and combinable. Omitting all three marks everything as read.
Stars or unstars an item.
Trigger an immediate poll for a feed.
Guided workflow prompt for research: search for a topic, explore similar items, and build a collection.
Full-text search items by title, content, or note. Returns paginated results with excerpts.
Subscribe to a new feed by URL.
Applies or removes a tag on an item. action is 'add' or 'remove'.
Guided workflow prompt for weekly digest: review starred items and topics from the past week.
Parameter constraint details missing from descriptions. 'limit' parameters lack explicit bounds (1-200) in descriptions; LLMs must infer from code. 'action' in tag_item lacks enum values in description.
Error handling lacks recovery guidance. ToolError exceptions are raised but descriptions do not suggest next steps (e.g., 'Feed not found: {feed_id}' should suggest 'Try list_feeds() to see available feeds').