KnowledgeGraphBLOCK
MCP server for enabling persistent knowledge storage for Claude through a knowledge graph with multiple storage backends and fuzzy search
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](https://mseep.ai/app/n-r-w-knowledgegraph-mcp)
I've become disillusioned with automated context management tools like this, as it's nearly impossible to control. After a while, I always have to manually clean up the mess or correct inappropriate LLM notes. Instead, I created a tool that gives the LLM agent access to dynamically loaded context. However, the context itself is user-created: https://github.com/n-r-w/agent-standards-mcp
KnowledgeGraph MCP Server
A simple way to give LLMs persistent memory across conversations. This server lets Claude or vscode remember information about you, your projects, and your preferences using a knowledge graph.
Key Features:
- Multiple Storage Backends: PostgreSQL (recommended) or SQLite (local file)
- Project Separation: Keep different projects isolated (auto-detected using prompts)
- Better Search: Find information with fuzzy search and pagination
Complete Setup Guide
Follow these steps in order to get the knowledge graph working with Claude:
Step 1: Choose Your Installation Method
Option A: NPX (Easiest - No download needed)
# Test that it works npx knowledgegraph-mcp --help
Option B: Docker
# Clone and build git clone https://github.com/n-r-w/knowledgegraph-mcp.git cd knowledgegraph-mcp docker build -t knowledgegraph-mcp .
Step 2: Choose Your Database
SQLite (Default - No setup needed):
- No database installation required
- Database file created automatically in
[you home folder]/.knowledge-graph/ - Perfect for personal use and most scenarios
- This is the default backend
PostgreSQL (For advanced users):
- Install PostgreSQL on your system
- Create a database:
CREATE DATABASE knowledgegraph; - Better for production use with multiple concurrent users
Step 3: Configure client
Claude Desktop
Edit your Claude Desktop configuration
270414469cf1OBSERVED · 2026-10-09Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add knowledgegraph-mcp -- npx -y [email protected]
{
"mcpServers": {
"knowledgegraph-mcp": {
"command": "npx",
"args": [
"-y",
"[email protected]"
]
}
}
}Exposed tools (10)
2 read · 4 write · 4 destructive. Blast radius: 4 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.
| Tool | Risk | Description |
|---|---|---|
add_observations | write | ADD new factual observations to existing entities. Requires exact entity names that exist in the knowledge graph. Use search_knowledge first to verify entity existence. |
add_tags | write | ADD categorical TAGS to existing entities for filtering and organization. Tags are case-sensitive exact-match labels used for quick retrieval with search_knowledge. Common categories: status, priority, type, domain. |
create_entities | write | CREATE new entities with OBSERVATIONS and optional tags. MANDATORY: If creating multiple entities, use a SINGLE call (batch). Each entity requires a unique name, type, and at least one observation. Ignores existing names. Use search_knowledge first. |
create_relations | write | CREATE directional RELATIONSHIPS between existing entities. Both source and target entities must exist. Use active voice relationship types (e.g., |
delete_entities | destructive | Permanently DELETE entities and all their RELATIONSHIPS. This action is irreversible and cascades to remove all connections. Verify entity existence with search_knowledge first. Consider delete_observations for partial updates. |
delete_observations | destructive | DELETE specific OBSERVATIONS from entities while preserving the entity, its relationships, and tags. Use for correcting errors or removing outdated facts without deleting the entire entity. |
delete_relations | destructive | DELETE specific RELATIONSHIPS between entities while preserving both entities. Use for updating connection status when relationships change (job changes, project completion, etc.). |
open_nodes | read | RETRIEVE specific ENTITIES by exact names along with their interconnections. Returns detailed information about the specified entities and relationships between them. Requires knowing exact entity names. |
read_graph | read | RETRIEVE the complete KNOWLEDGE GRAPH with all entities and relationships for a project. Returns comprehensive view of the entire network structure. Can be large for projects with many entities. |
remove_tags | destructive | REMOVE specific TAGS from existing entities to maintain accurate categorization. Use for status updates, priority changes, or cleaning up outdated classifications. Tags must match exactly (case-sensitive). |
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (6 observation(s))
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (14)
# For PostgreSQL testing: postgresql://postgres:password@localhost:5432/knowledgegraph_test
"KNOWLEDGEGRAPH_CONNECTION_STRING": "postgresql://postgres:yourpassword@localhost:5432/knowledgegraph"
"-e", "KNOWLEDGEGRAPH_CONNECTION_STRING=postgresql://postgres:[email protected]:5432/knowledgegraph",
"-e", "KNOWLEDGEGRAPH_CONNECTION_STRING=postgresql://postgres:[email protected]:5432/knowledgegraph",
KNOWLEDGEGRAPH_TEST_CONNECTION_STRING="postgresql://postgres:password@localhost:5432/knowledgegraph_test"
delete_entities, delete_observations, delete_relations, remove_tags
import { Entity } from '../../core.js';import { Entity } from '../../core.js';import { Entity } from '../../core.js';import { KnowledgeGraph, Entity, Relation } from '../../index.js';import { KnowledgeGraph, Entity, Relation } from '../../core.js';better-sqlite3, fuse.js, pg, @types/better-sqlite3, @types/jest, @types/node, @types/pg, jest
- **Current project entities**: Full access for modifications
- **Current project entities**: Full access for modifications
Gates applied: no_behavioural_pass.
270414469cf1full audit observations/trust-audit/mcp-server/n-r-w__knowledgegraph.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-09 | 270414469cf1 | BLOCK | D | 69 | first audit |
Questions
What is the KnowledgeGraph MCP server?
MCP server for enabling persistent knowledge storage for Claude through a knowledge graph with multiple storage backends and fuzzy search
What tools does KnowledgeGraph expose?
10 in total: 2 read-only, 4 that write, and 4 that can delete or overwrite (delete_entities, delete_observations, delete_relations, remove_tags). Every one is listed on this page with its risk.
Is KnowledgeGraph safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (69/100) and found 1 critical or high issue in the source. Each one is listed on this page with the file and line it is on. Separately from the audit: 4 of its tools can destroy data, so scope the token you give it to what you actually need.
What credentials does KnowledgeGraph need?
No credential environment variables were found in its source, so it appears to need none.
How does KnowledgeGraph run?
It speaks stdio, so it runs as a local process your client starts. It is published on npm as knowledgegraph-mcp at 1.7.4.
How current is this page?
The grade is for one exact copy of the source (270414469cf1), read on 2026-10-09. The repository is watched and re-audited when it changes.