LightRAG MCP
Server:
user-daniel_lightrag_mcp/daniel_lightrag_mcp
Underlying service: Serve · LightRAG
Cross-session knowledge base for P0 retrieval, P2.5 storage, and P3 milestone ingestion.
Daily Tools
| Tool | Stage | Notes |
|---|---|---|
query_text | P0 | mode: hybrid; default only_need_context: true |
insert_text | P2.5/P3 | Prefix: [tags: …] [type: …] [date: …] |
insert_texts | P2.5 | Batch; each item: title + content + optional metadata |
get_track_status | P2.5/P3 | Required for milestone ingestion confirmation |
check_entity_exists | P2.5 | Optional before "remember / update decision" |
query_text Parameters
| Param | Recommended | Notes |
|---|---|---|
query | Keywords from user input | Natural-language question |
mode | hybrid | Entity-local: local; global: global; simple: naive |
only_need_context | true (must pass explicitly) | Context only; pass false when user wants a direct answer |
insert_text Body Format
[tags: decision,api] [type: decision] [date: 2026-06-10]
<decision body>Milestone Tags
| Must track | May skip track |
|---|---|
decision, api, preference, plan, ui, architecture | brainstorming, debug, note, wip |
upload_document always requires track.
On-Demand Tools
| Tool | When |
|---|---|
upload_document | File ingestion |
get_documents / get_documents_paginated | Browse documents |
get_document_status_counts | Ops statistics |
get_knowledge_graph / get_graph_labels | Structure review |
scan_documents | Trigger scan |
get_health | After 2 consecutive failures; not every session |
update_entity / update_relation | Graph edits (need ID); use new insert_text for body changes |
delete_* | Ops cleanup (user confirmation required) |
query_text_stream | Long-answer scenarios |
Explicit Trigger (Hard Gate)
User message contains knowledge base / LightRAG / daniel_lightrag → first step query_text, before reading code, P1, or implementation.
Failure Diagnosis
query_text / insert_text fails twice in a row → get_health → if still failing, skip knowledge-base steps.
Examples
text
Implement user login API per knowledge base standardstext
Remember: project uses PostgreSQL on port 5432