loci-mcp 0.2.3

Cognitive memory MCP server — persistent, structured, cross-session memory for AI agents
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35

# Loci Configuration
# Copy to ~/.loci/config.toml and modify as needed.
# All values shown are defaults.

[server]
transport = "stdio"                       # "stdio" | "sse" (SSE is M7+)
log_level = "info"                        # "error" | "warn" | "info" | "debug" | "trace"

[storage]
db_path = "~/.loci/memory.db"             # Path to SQLite database
default_group = "default"                 # Default memory group

[embedding]
provider = "local"                        # "local" | "voyage" | "openai"
model = "all-MiniLM-L6-v2"               # ONNX embedding model name
cache_dir = "~/.loci/models"              # Directory for cached model files

[retrieval]
default_max_results = 5                   # Max results per recall_memory call
preload_token_budget = 2000               # Token budget for preloaded context
recall_token_budget = 4000                # Token budget for recall responses
rrf_k = 60                               # Reciprocal Rank Fusion k parameter
dedup_threshold = 0.92                    # Cosine similarity threshold for deduplication

[maintenance]
enabled = false                           # Enable automatic maintenance (future M7)
interval_days = 7                         # Days between maintenance cycles
episodic_decay_factor = 0.95              # Confidence multiplier per cycle (episodic)
semantic_decay_factor = 0.99              # Confidence multiplier per cycle (semantic/procedural/entity)
compaction_age_days = 30                  # Episodic memories older than this are compaction candidates
compaction_min_group_size = 5             # Minimum memories in a week+group to trigger compaction
promotion_threshold = 3                   # Similar episodics needed to promote to semantic
promotion_similarity = 0.88              # Cosine similarity threshold for promotion clustering
cleanup_confidence_floor = 0.05           # Memories below this confidence are cleanup candidates
cleanup_no_access_days = 90               # Days without access before cleanup eligibility