Every series memors-mcp exposes on master: v1.4 plus the unreleased section of the changelog. Types: C counter, G gauge, H histogram.
A histogram exposes _bucket{le}, _sum and _count. A family with no observations yet
shows only its # HELP and # TYPE lines.
Bucket sets:
Set Boundaries
latency (s) 0.001, 0.0025, 0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10 tokens 50, 100, 200, 500, 1000, 2000, 4000, 8000, 16000 count 0, 1, 2, 5, 10, 20, 50, 100, 200, 500 bytes 1 KiB, 4 KiB, 16 KiB, 64 KiB, 256 KiB, 1 MiB, 4 MiB
Metric Type Labels Meaning
memors_mcp_requests_totalC method, outcome = ok, errorEvery MCP request received (tools/call, resources/read, server/discover, …) memors_mcp_requests_in_flightG — Requests being served now memors_mcp_tool_calls_totalC tool, outcome = ok, tool_error, input_required, errorTool calls. input_required is promote waiting for a human memors_mcp_tool_call_duration_secondsH latency toolTool call latency memors_mcp_tool_result_tokensH tokens toolEstimated tokens in each tool result: the cost the agent pays memors_mcp_tool_errors_totalC tool, class = not_found, forgotten, needs_human, canceled, invalid_args, internalWhy calls failed memors_mcp_elicitations_totalC tool, outcome = asked, accept, decline, cancel, unsupportedHuman approval dialogs memors_mcp_resource_reads_totalC kind = doc, chunk, source, fact, entity, page, index, ns_index, unknown; outcome = ok, errormemo:// resource readsmemors_mcp_prompt_gets_totalC prompt = ask, unknown; outcome = ok, errorPrompts fetched by clients memors_mcp_sessions_totalC clientSessions started, by client name
tool is one of the ten tool names, or unknown when a client calls a tool the server does
not have; such a call has outcome="error". Clients cannot add label values.
Metric Type Labels Meaning
memors_search_totalC mode_requested, mode_resolved, granularity, outcome = results, abstainSearches that completed (failed searches show up as tool errors) memors_search_duration_secondsH latency granularityEnd-to-end search latency memors_search_arm_duration_secondsH latency armTime in each retrieval arm memors_search_arm_candidatesH count armCandidates each arm returned before fusion memors_search_resultsH count granularityResults returned per search memors_search_truncated_results_totalC — Ranked results left out by the token budget memors_search_cutoff_totalC kind = gap, budget, limit, noneWhy lists ended memors_search_degraded_totalC reason = no_embedder, query_embed_failed, reranker_failed, otherSearches served without a capability memors_search_abstentions_totalC reason = no_match, no_arm, no_fact, no_page, excluded, otherSearches that returned nothing on purpose. excluded: every match was in exclude_ids memors_search_entities_matchedH count — Known entities named per query memors_search_rerank_duration_secondsH latency modelCross-encoder time, only with the reranker memors_graph_cache_totalC event = hit, build, build_asofIn-memory mention-graph cache memors_graph_build_duration_secondsH latency — Time to build a namespace graph
Metric Type Labels Meaning
memors_store_writes_totalC op, channel = tool, cli, elicitation, workerEvery audited write memors_store_ingests_totalC outcome = new, revision, dedup, errorIngest calls memors_store_ingest_duration_secondsH latency — Ingest latency, embedding included memors_store_chunks_written_totalC — Passages written memors_store_document_bytesH bytes — Size of ingested documents memors_store_vectors_stored_totalC modelVectors stored memors_store_embed_batches_totalC outcome = ok, error, queuedEmbedding batches on the write path memors_store_jobs_totalC kind, event = queued, done, failedBackground jobs, for example the embed backlog memors_store_mentions_linked_totalC — Entity mentions linked to passages memors_store_pages_marked_stale_totalC — Pages marked stale by a source change or forget memors_store_work_items_totalC kind, event = created, done, skippedCompaction work items memors_store_status_scrape_errors_totalC — Failures reading the gauges below
Metric Labels Meaning
memors_kb_sources— Sources memors_kb_documents_live— Live documents (latest revision, not forgotten) memors_kb_revisions— Document revisions stored memors_kb_chunks— Live passages memors_kb_facts— Live facts memors_kb_entities— Entities memors_kb_mentions— Entity mentions memors_kb_edges— Typed edges memors_kb_merge_review_open— Open merge candidates awaiting a human memors_kb_pages— Curated pages memors_kb_pages_stale— Stale pages memors_kb_work_items_open— Open compaction work items memors_kb_jobs_queued— Jobs queued or running memors_kb_jobs_failed— Jobs failed memors_kb_pending_embeddingsmodelPassages without a vector for that model memors_kb_schema_version— Schema version of the file memors_kb_db_size_bytes— Size of the main SQLite file, excluding the WAL memors_kb_last_write_timestamp_seconds— Unix time of the last audited write memors_kb_namespace_documentsnamespaceLive documents per namespace memors_kb_namespace_chunksnamespaceLive passages per namespace
All are gauges.
Metric Type Labels Meaning
memors_embed_duration_secondsH latency model, role = query, docEmbedding call latency memors_embed_texts_totalC model, roleTexts embedded memors_embed_errors_totalC model, roleEmbedding calls that failed memors_embed_batch_sizeH count modelTexts per call memors_embed_model_downloads_totalC model, outcome = ok, error, corrupt_retryModel downloads memors_embed_model_load_secondsG modelTime the last model load took memors_embed_model_loadedG model, backend = hugot, static1 while a model is loaded in this process
Metric Type Labels Meaning
memors_ui_requests_totalC route (for example GET /search), statusUI requests memors_ui_request_duration_secondsH latency routeUI render time
Metric Type Labels Meaning
memors_build_infoG (always 1) version, go_version, mcp_protocol, goos, goarchWhat is running go_infoG versionGo version go_goroutinesG — Goroutines go_memstats_heap_alloc_bytesG — Live heap go_memstats_sys_bytesG — Memory obtained from the OS go_gc_cycles_totalC — Completed GC cycles go_gc_pauses_secondsH — GC pause distribution process_start_time_secondsG — Process start time; use it to spot session restarts