Graph-Mem

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/sathvik-1007-graphmem-mcp

About

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.

Tools

12 read · 8 write · 4 delete · 5 not yet classified

Secrets

None asked for

Install scripts

Not scanned yet

Runs as

Python package

Tools (29)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

• add_entities

  Add entities to the knowledge graph. Entities with the same name and type are automatically merged. Each entity needs: name (str), entity_type (str, e.g. 'person', 'concept', 'place'). Optional: description (str), properties (dict), observations (list[str]).

• add_observations

  Add factual observations to an existing entity. Observations are atomic statements about an entity (facts, quotes, events). They are embedded separately for fine-grained semantic search. Args: entity_name: Name of the entity to attach observations to. observations: List of observation text strings. source: Optional provenance string (e.g. session ID, document name).

• add_relationships

  Add relationships (edges) between entities in the knowledge graph. Each relationship needs: source (str, entity name), target (str, entity name), relationship_type (str, e.g. 'knows', 'works_at', 'depends_on'). Optional: weight (float, 0-1, default 1.0), properties (dict). Duplicate edges (same source, target, type) are merged with the higher weight kept.

• audit_graph

  Name every entity with a data-quality problem — the thorough, expensive check to run after a bulk import or before a cleanup pass. Unlike read_graph (shape only) and graph_health (fixed-cost signals plus advice), this scans up to 2000 entities and lists the offenders in five categories: disconnected, missing description, missing observations, empty properties, and only-one-relationship. Returns a dict; ``report`` holds the same findings rendered as plain text for direct reading, and ``truncated`` says whether the graph outgrew the scan limit.

• compact_observations

  Atomic observation compaction — delete old observations and add new ones in one step. Use this to consolidate many observations into fewer, denser ones. You decide what to merge — this tool makes it atomic so no data is lost. Workflow: get_entity → read observations → decide which to keep vs merge → call compact_observations with keep_ids (observations to preserve unchanged) and new_observations (your merged/summarized replacements). Args: entity_name: Name of the entity to compact. keep_ids: Observation IDs to preserve unchanged. All others are deleted. Duplicates are count…

• create_graph

  Create a new empty knowledge graph. Creates a new .db file in the .graphmem/ directory. The graph can then be activated with switch_graph. Args: name: Name for the new graph. Letters, digits, hyphens, and underscores only, at most 64 characters. Cannot be 'graph' (reserved for the default graph's file — use 'default').

• delete_entities

  Remove entities from the knowledge graph by name. Cascades to delete all related observations and relationships. Returns the count of entities actually deleted.

• delete_graph

  Delete a knowledge graph permanently. Removes the .db file and associated WAL/SHM files from the .graphmem/ directory. Cannot delete the currently active graph — switch to a different graph first. Args: name: Name of the graph to delete. Cannot be the active graph.

• delete_observations

  Remove specific observations from an entity by observation ID. Validates that the observations belong to the specified entity before deleting. Also cleans up associated embeddings. Args: entity_name: Name of the entity the observations belong to. observation_ids: List of observation IDs to delete.

• delete_relationships

  Remove relationships between two entities by name. Deletes matching edges from source to target. If relationship_type is provided, only relationships of that type are deleted. Otherwise all relationships between the pair are removed. Args: source: Source entity name. target: Target entity name. relationship_type: Optional — restrict deletion to this edge type.

• find_connections

  Start from ONE entity and list everything reachable from it — use this to answer "what is related to X?". Differs from get_subgraph (many seeds, returns entities *and* the edges between them, for visualising a region) and from find_paths (two fixed endpoints, returns the routes between them). This one returns a flat, depth-ordered list with one shortest route to each entity found. Args: entity_name: Starting entity name. max_hops: How far to traverse (clamped to 1-10, default 2). relationship_types: Optional filter on edge types (e.g. ['knows', 'works_at']). direction: 'outgo…

• find_paths

  Show how TWO named entities connect — use this when you already know both endpoints and want the chain between them. Differs from find_connections (one seed, everything reachable) and from get_subgraph (a region, not a route). Returns up to 10 shortest paths, each a sequence of entities and relationship types. Args: source: Starting entity name. target: Destination entity name. max_hops: Maximum path length to consider (clamped to 1-10, default 5).

• get_entity

  Get full details of a single entity by name, including observations and relationships. Uses fuzzy name resolution: exact match -> case-insensitive -> FTS5 suggestions. At most 50 observations (newest first) and 50 relationships are attached; ``observations_truncated`` / ``relationships_truncated`` say when more exist, and ``observation_count`` / ``relationship_count`` give the true totals. Use search_observations scoped by entity_name to reach the rest.

• get_subgraph

  Extract a whole region around SEVERAL seed entities — use this when you need the edges too, e.g. to draw or reason about a neighbourhood. Differs from find_connections (single seed, flat list of reachable entities, no edge list) and from find_paths (routes between two named endpoints). Args: entity_names: Seed entity names to expand from. radius: How many hops to expand (clamped to 1-5, default 2). Returns: ``{entities, relationships, truncated}``. At most 200 entities are returned, with relationships restricted to that set; ``truncated`` is true when the region was larg…

• graph_health

  Ask "does this graph need cleaning?" — returns maintenance signals and a short list of suggested actions, at fixed cost regardless of graph size. Sits between the other two overview tools: read_graph describes the graph's shape and says nothing about quality; audit_graph names every problem entity and scans the whole graph to do it. This one reports counts, the top 5 observation hotspots, up to 10 entities missing descriptions, the top 10 entity types, and what to do about them.

Signals

Stars

1

Forks

0

Open issues

0

Weekly downloads

31

Last commit

Jul 21, 2026 (3 months ago)

Last release

Jul 21, 2026 (3 months ago)

Last package publish

Jul 21, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026) and the package registry.

Details

Registry name

io.github.Sathvik-1007/graphmem-mcp

Version

1.0.0

Auth

No auth

Packages

pypi graphmem-mcp

License

MIT

Repository

github.com/Sathvik-1007/GraphMem-MCP

Website

github.com/Sathvik-1007/GraphMem-MCP#readme

First seen

Oct 24, 2026

Last checked

Oct 24, 2026

Source

Official MCP Registrygithub-graphql