Local-first cross-corpus retrieval MCP server — model2vec embeddings + LanceDB (Tantivy BM25) + SQLite. One binary; hybrid (dense + BM25) search across markdown, code, and Claude Code sessions.
Drift inferred · capture-to-capture
No drift recorded — single capability capture; advisories appear once its surface changes.
tools
-
attention_attend
{ scope?, context } — ingest context into the scope's attention vector
-
attention_decay
{ handle, factor } — multiplicative fade on the floor
-
attention_familiarize
{ scope?, handle } — increment the familiarity counter
-
attention_focus
{ scope?, query, surface_limit?: 0..100 } — pin a focus query
-
attention_fold
{ scope?, handle, depth: folded\
-
attention_refocus
{ scope?, query, surface_limit? } — rotate to a query already in focus history
-
attention_status
{ scope?, surface_limit? } — read-only snapshot of pin / history / transient
-
attention_surface
{ scope?, limit?: 1..200 } — pages above the archive threshold
-
attention_unfocus
{ scope?, surface_limit? } — clear the pin
-
memory_concept
{ action: show\
-
memory_connect
{ from, relation, to, project?, evidence? } — author a concept edge
-
memory_focus
{ target, project? } — pin a focus that biases recall
-
memory_recall
{ query, project?, limit?, source?, learn? } — recall, optionally learning concept candidates
-
memory_reflect
{} — consolidation pass: promote corroborated candidates, re-resolve evidence, and promote off-diagonal latent bridges into weak promoted edges
-
memory_remember
{ kind: concept_seed\
-
memory_surface
{ view?: now\
-
recall
{ query, project?, source?, since?, limit?: 1..100 }
-
recall_audit
{ sql } — single read-only SELECT over the audit_events table
-
recall_fault
{ query, intent?: symbol\
-
recall_link
{ chunk_id } — the chunk plus its parent chain
-
recall_stats
{} — total count, breakdown by source, model info, last scan
-
thread_attention
{ since_hours?, limit?, samples_per_burst?, decay_hours? } — activity-burst surface
-
thread_create
{ scope?, handle, body?, tension?: active\
-
thread_emergent
{ since_hours?, limit?, min_cluster_size?, cohesion_threshold?, ... } — embedding-density clusters
-
thread_evidence
{ action: add\
-
thread_link
{ scope?, handle, target_path, category }
-
thread_list
{ scope?, tension? }
-
thread_novelty
{ since_hours?, baseline_days?, limit?, min_mean_novelty?, ... } — divergence-from-baseline clusters
-
thread_promote
{ handle_from_proposed, target_tier: active\
-
thread_query
{ signals?: [density\
-
thread_unlink
{ evidence_id }
-
tool
input (one-line)
This server hasn't been statically analyzed yet (npm/PyPI artifacts only).