raggy-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| RAGGY_API_KEY | No | API key for Pro tier access. If not set, free tier (20 searches/day) is used. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| raggy_uploadA | Upload files, URLs, or long-form content as a private knowledge base source. Use for documents, articles, or reference material the user wants indexed and searchable. For short notes/thoughts use raggy_capture instead. Requires API key. |
| raggy_private_sourcesA | List uploaded private knowledge base sources (files/URLs ingested via raggy_upload). Use when the user asks "what sources have I uploaded" or to find a source ID for deletion. Requires API key. |
| raggy_delete_sourceA | Delete an uploaded private knowledge base source by ID. Use when the user asks to remove a specific source. For individual memories use raggy_forget instead. Get the ID via raggy_private_sources. Requires API key. |
| raggy_rememberA | Save a simple, unstructured note. Prefer raggy_capture for anything with a clear type (decision/error/insight/snippet). Only use raggy_remember for free-form notes where metadata adds no value. |
| raggy_recallA | Targeted semantic search over memories. Only use AFTER raggy_context at session start has returned nothing relevant, or for a specific follow-up lookup mid-session. Do NOT call twice per question. For time-based queries use raggy_timeline. |
| raggy_captureA | Auto-capture structured memory. Use for: decisions (content_type="decision"), errors+fixes ("error"), preferences ("note" + tags:["preference"]), insights ("insight"), snippets ("snippet"), research ("research"). Set importance: high for decisions/errors/prefs, medium for insights/snippets. Always include source_app + capture_context.agent_id. |
| raggy_contextA | MANDATORY FIRST ACTION of every session. Call ONCE before answering the first user message with their initial request as the query. Loads relevant memories from prior sessions. Do not make a second call unless this returned empty and you need a targeted lookup (then use raggy_recall). |
| raggy_timelineA | Use for time-based queries like "what did we work on today/yesterday/last week". Pass since as a relative window (e.g. "24h", "7d") or ISO date. Do NOT use raggy_recall for temporal questions — this is the right tool. |
| raggy_linkA | Connect two related memories in the knowledge graph. Use when a new memory supersedes/refines/contradicts an old one, or when an error resolves a past bug. Relationship types: caused_by, resolved_by, supersedes, refines, contradicts, related_to, follows_from, part_of. |
| raggy_forgetA | Call immediately when the user says "forget that", "don't save that", or "delete that memory". Use mode="delete" to remove entirely or mode="redact" to preserve the graph node but strip content. Not for uploaded sources — use raggy_delete_source for those. |
| raggy_threadsA | Use for session-based queries like "what was in my last session" or "what did we discuss in that conversation about X". action="list" browses recent sessions; action="get" retrieves all memories for one session_id. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 11 tools
Tools are largely distinct, but raggy_remember and raggy_capture both create memories, and raggy_context and raggy_recall both handle memory retrieval. Descriptions do separate them well via structured vs unstructured capture and session-start vs targeted lookup, so confusion is unlikely in practice.
All tools share the raggy_ prefix and snake_case formatting, but naming style is mixed: upload, remember, capture, recall, forget, and link are verbs, while context, timeline, threads, and private_sources are noun phrases. There is no consistent verb_noun pattern.
11 tools is well-scoped for a memory/knowledge server covering ingestion, retrieval, deletion, graph linking, and session introspection. Each tool has a clear role and the count feels appropriate rather than bloated or thin.
The set covers upload/list/delete for private sources, capture/remember/forget for memories, context/recall/timeline/threads for retrieval, and link for graph relationships. The main gap is no explicit update/edit operation for a memory, though supersedes links and forget redaction partially address that.