本文へ移動
cccskills
無料GitHub で公開

neo4j-mcp-skill

Use when installing, configuring, or troubleshooting the official Neo4j MCP server (neo4j/mcp) — connecting Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Kiro, or other MCP-compatible editors to a Neo4j database via stdio or HTTP transport. Covers the four MCP tools (get-schema, read-cypher, write-cypher, list-gds-procedures), read-only mode, and multi-database configuration. Does NOT cover writing Cypher queries via those tools — use neo4j-cypher-skill. Does NOT cover agent memory — use neo4j-agent-memory-skill. Does NOT cover Aura instance provisioning — use neo4j-aura-provisioning-skill.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md11.6 KB
  • README.md517 B

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Neo4j MCP Skill

Installs and configures the official Neo4j MCP server so AI agents can connect to Neo4j via any MCP-compatible client.

When to Use

  • Connecting Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Kiro, or another editor to Neo4j via MCP
  • Installing neo4j-mcp-server and writing the correct config for a specific editor
  • Switching between stdio and HTTP transport
  • Enabling or disabling write access (NEO4J_READ_ONLY)
  • Troubleshooting "MCP server not found" or connection errors

When NOT to Use

  • Writing or optimizing Cypher queries → use neo4j-cypher-skill
  • Provisioning a new Neo4j Aura instance → use neo4j-aura-provisioning-skill
  • Agent long-term memory → use neo4j-agent-memory-skill
  • neo4j-admin / cypher-shell / aura-cli → use neo4j-cli-tools-skill

Available MCP Tools

ToolTypeWhat it does
get-schemareadReturns labels, relationship types, property keys, and indexes
read-cypherreadExecutes read-only Cypher (MATCH, RETURN, SHOW)
write-cypherwriteExecutes write Cypher (MERGE, CREATE, SET, DELETE) — disabled when NEO4J_READ_ONLY=true
list-gds-proceduresreadLists available Graph Data Science procedures (requires GDS plugin)

Installation

Step 1 — Install and find the absolute path

# Option A: pip (recommended)
pip install neo4j-mcp-server

# Option B: Download binary
# https://github.com/neo4j/mcp/releases -- macOS, Linux, Windows binaries

# Option C: Docker
docker pull neo4j/mcp

Get the absolute path — you will need this in Step 3:

which neo4j-mcp          # e.g. /usr/local/bin/neo4j-mcp
                         # or:  /Users/you/project/.venv/bin/neo4j-mcp  (if installed in venv)

neo4j-mcp --version      # confirm it runs

Why absolute path matters: editors (Claude Code, Cursor, Claude Desktop) spawn the MCP server as a subprocess using their own restricted PATH — not your shell's PATH. On macOS, GUI apps do not inherit .zshrc or .zprofile. Using neo4j-mcp as the command will silently fail; using /full/path/to/neo4j-mcp always works. Always use the output of which neo4j-mcp in the command field below.

Step 2 — Prepare credentials

# .env (gitignored)
NEO4J_URI=neo4j+s://<instance>.databases.neo4j.io   # Aura
# or bolt://localhost:7687                           # local
NEO4J_USERNAME=neo4j
NEO4J_PASSWORD=<password>
NEO4J_DATABASE=neo4j

Verify connectivity before configuring the editor:

source .env
cypher-shell -a "$NEO4J_URI" -u "$NEO4J_USERNAME" -p "$NEO4J_PASSWORD" \
  "RETURN 'connected' AS status"
# If cypher-shell not available: python3 -c "
# from neo4j import GraphDatabase, __version__
# d = GraphDatabase.driver('$NEO4J_URI', auth=('$NEO4J_USERNAME','$NEO4J_PASSWORD'))
# d.verify_connectivity(); print('connected'); d.close()"

Step 3 — Configure your editor

Pick the config block for your editor. All use STDIO transport (the MCP server runs as a subprocess of the editor).

Claude Code — add to ~/.claude/settings.json. If the file already exists, merge the neo4j block into the existing mcpServers object — do not replace the whole file.

{
  "mcpServers": {
    "neo4j": {
      "command": "/full/path/to/neo4j-mcp",
      "env": {
        "NEO4J_URI": "neo4j+s://<host>",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "<password>",
        "NEO4J_DATABASE": "neo4j",
        "NEO4J_READ_ONLY": "true"
      }
    }
  }
}

CLI alternative (sets command only — you still need to add env vars to the file):

claude mcp add neo4j /full/path/to/neo4j-mcp

Claude Desktop

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "neo4j": {
      "command": "/full/path/to/neo4j-mcp",
      "env": {
        "NEO4J_URI": "neo4j+s://<host>",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "<password>",
        "NEO4J_DATABASE": "neo4j",
        "NEO4J_READ_ONLY": "true"
      }
    }
  }
}

VS Code — .vscode/mcp.json (note: uses servers, not mcpServers — different from all other editors):

{
  "servers": {
    "neo4j": {
      "type": "stdio",
      "command": "/full/path/to/neo4j-mcp",
      "env": {
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "password",
        "NEO4J_DATABASE": "neo4j",
        "NEO4J_READ_ONLY": "true"
      }
    }
  }
}

Cursor — global: ~/.cursor/mcp.json / project: .cursor/mcp.json (same structure as Claude Code, uses mcpServers).

Windsurf — global: ~/.codeium/windsurf/mcp_config.json / project: .windsurf/mcp_config.json (same structure as Claude Code, uses mcpServers).

Kiro — global: ~/.kiro/settings/mcp.json / project: .kiro/settings/mcp.json. Supports ${VARIABLE} to pull from the shell environment exported before the editor launched:

{
  "mcpServers": {
    "neo4j": {
      "command": "/full/path/to/neo4j-mcp",
      "env": {
        "NEO4J_URI": "${NEO4J_URI}",
        "NEO4J_USERNAME": "${NEO4J_USERNAME}",
        "NEO4J_PASSWORD": "${NEO4J_PASSWORD}",
        "NEO4J_DATABASE": "neo4j",
        "NEO4J_READ_ONLY": "true"
      }
    }
  }
}

Cline — ~/.vscode/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json (same structure as Claude Code, uses mcpServers). Or add via VS Code settings → Cline extension → MCP Servers panel.

Antigravity — mcp_config.json at project root (same structure as Claude Code, uses mcpServers).

Step 4 — Restart the editor

After editing the config file, restart the editor (or reload the MCP server if the editor supports hot-reload). Verify the server appears in the editor's MCP panel or tool list.

Step 5 — Smoke test

Run get-schema via the agent or directly via MCP. Do not use "does it return labels?" as the test — an empty database returns no labels and looks identical to a broken connection.

Use this instead — it succeeds even on an empty DB:

read-cypher: RETURN 'connected' AS status

Expected: { "status": "connected" }. Any result confirms the server is alive and the credentials are valid.

If the database has data, also run:

get-schema

and confirm it returns the node labels and relationship types you expect.


HTTP Transport

HTTP transport runs the MCP server as a persistent network service — useful for shared servers, containers, or multiple clients.

# With credentials baked in (simpler — server authenticates all clients as the same user)
neo4j-mcp \
  --neo4j-transport-mode http \
  --neo4j-http-host 127.0.0.1 \
  --neo4j-http-port 8080 \
  --neo4j-uri bolt://localhost:7687 \
  --neo4j-username neo4j \
  --neo4j-password <password> \
  --neo4j-database neo4j

Editor config for HTTP (no credentials in config — server holds them):

{
  "mcpServers": {
    "neo4j-http": {
      "type": "http",
      "url": "http://127.0.0.1:8080/mcp"
    }
  }
}

Per-request auth (omit --neo4j-username/--neo4j-password from the server command): each client request must include an Authorization: Basic <base64(username:password)> header. Generate the value with:

echo -n "neo4j:<password>" | base64

Then add to the editor config:

{
  "headers": { "Authorization": "Basic bmVvNGo6cGFzc3dvcmQ=" }
}

Security: bind to 127.0.0.1, not 0.0.0.0. Only change to a broader interface if you need remote access and have TLS + auth in front of it.


Environment Variables

VariableRequiredDefaultNotes
NEO4J_URIyes—neo4j+s:// for Aura; bolt:// for local
NEO4J_USERNAMEyes—
NEO4J_PASSWORDyes—
NEO4J_DATABASEnoserver defaultSpecify explicitly to avoid routing to wrong DB
NEO4J_READ_ONLYnofalseSet true to hide write-cypher (safe default for exploration)

When to set NEO4J_READ_ONLY=false (or remove the variable): any use case that needs to write data — imports, MERGE, schema changes, GDS write-back. Without write access, the agent will see only three tools (get-schema, read-cypher, list-gds-procedures) and write-cypher calls will fail.


APOC Requirement

The MCP server uses APOC for schema introspection (get-schema). On Aura, APOC is included automatically. On self-managed Neo4j, install the APOC plugin and whitelist it:

# neo4j.conf
dbms.security.procedures.unrestricted=apoc.*

Verify: read-cypher: RETURN apoc.version() AS v — if this fails, get-schema will return incomplete results.


Troubleshooting

SymptomLikely causeFix
Server not listed in editorConfig file path wrong, JSON malformed, or editor not restartedValidate JSON (python3 -m json.tool settings.json); confirm correct file for your editor; restart editor
neo4j-mcp: command not foundEditor PATH doesn't include the binary locationUse absolute path in command: run which neo4j-mcp in terminal and paste the result
Server starts then immediately exitsPython not found (pip install) or wrong binary for OS (binary install)Run the command manually in terminal to see the error: /full/path/neo4j-mcp --version
AuthenticationExceptionWrong credentialsVerify URI + credentials with cypher-shell or driver before editing config
write-cypher tool missingNEO4J_READ_ONLY=trueRemove or set to false if writes are needed
ServiceUnavailableDB not reachable from editor processCheck firewall, VPN, or whether Aura instance is paused
list-gds-procedures returns emptyGDS not installed or Aura FreeGDS requires Aura Professional/Enterprise or self-managed with plugin

Checklist

  • which neo4j-mcp run and absolute path noted — this goes in command, not neo4j-mcp
  • Connectivity verified (RETURN 'connected') before editing editor config
  • Credentials not committed to git; .env in .gitignore
  • NEO4J_READ_ONLY set intentionally — true for exploration, false/absent for write use cases
  • NEO4J_DATABASE specified explicitly
  • Correct config file used for the target editor (VS Code uses servers; all others use mcpServers)
  • Config is valid JSON (validate with python3 -m json.tool <file> before saving)
  • Editor restarted after config change
  • Smoke test passed: read-cypher: RETURN 'connected' AS status returns a result
  • APOC available on self-managed: read-cypher: RETURN apoc.version()

References

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Authoritative reference for the neo4j-agent-memory Python package — a graph-native memory system for AI agents built on Neo4j — and for the hosted service (NAMS) at memory.neo4jlabs.com. Use this skill whenever the user mentions neo4j-agent-memory, agent memory with Neo4j, context graphs, the POLE+O model, MemoryClient/MemorySettings, the memory MCP server, or any of the framework integrations (LangChain, PydanticAI, CrewAI, AWS Strands, Google ADK, Microsoft Agent Framework, OpenAI Agents, LlamaIndex). Also use when the user mentions the hosted service at memory.neo4jlabs.com, NAMS, the Neo4j Agent Memory Service, the `nams_` API key prefix, or the hosted MCP endpoint. Also use when writing documentation, blog posts, tutorials, PRDs, or code samples for the project, when comparing agent memory approaches, or when positioning graph-native memory against vector-only approaches — even if the user doesn't explicitly name the package.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

Manages Neo4j Aura Agents via the v2beta1 REST API — create, list, get, update, delete, and invoke Aura agents backed by an AuraDB instance. Use when configuring Aura Agent tools (CypherTemplate, SimilaritySearch, Text2Cypher), setting system prompts, deploying agents to REST or MCP endpoints, invoking agents with natural language queries, or generating evaluation datasets (max 50 questions) for the Aura Agent evaluation feature. Covers OAuth2 auth, organization/project scoping, tool parameter schemas, and InvokeAgentResponse format. Does NOT cover AuraDB instance provisioning — use neo4j-aura-provisioning-skill. Does NOT cover vector index creation — use neo4j-vector-index-skill.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

Serverless Aura Graph Analytics (AGA) GDS Sessions — covers GdsSessions, AuraGraphDataScience, AuraAPICredentials, DbmsConnectionInfo, SessionMemory, get_or_create, remote graph projection with gds.graph.project.cypher and gds.graph.project.remote, gds.graph.project.native, gds.graph.construct, graphdatascience client 2.0 session endpoints, async compute and gds.jobs, AuraDB Cypher API memory/sessionId projection, algorithms, write-back, and session lifecycle. Use for AuraDB-connected, self-managed Neo4j, or standalone DataFrame/Spark session workloads. Does NOT cover the embedded GDS plugin on Aura Pro or self-managed Neo4j — use neo4j-gds-skill. Does NOT handle Cypher authoring — use neo4j-cypher-skill. Does NOT cover Snowflake Graph Analytics — use neo4j-snowflake-graph-analytics-skill.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

Provisions and manages Neo4j Aura instances via CLI (aura-cli v1.7+) or REST API. Use when creating, pausing, resuming, resizing, or deleting AuraDB Free/Professional/Business Critical/VDC instances; downloading credentials; scripting CI/CD pipelines; polling async status; or using the Terraform neo4j-labs/neo4jaura provider. Covers auth setup (client credentials OAuth2), credential lifecycle (download once — never recoverable), instance type selection, region codes, and Python provisioning scripts. Does NOT handle Cypher queries — use neo4j-cypher-skill. Does NOT cover Graph Data Science algorithms — use neo4j-gds-skill or neo4j-aura-graph-analytics-skill. Does NOT cover neo4j-admin/cypher-shell — use neo4j-cli-tools-skill.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

Use when working with Neo4j command-line tools — neo4j-cli (modern unified CLI — Cypher via Bolt, schema inspection, Aura management, Docker containers, credential management, agent skill catalog install; install with curl -sSfL https://neo4j.sh/install.sh | bash), neo4j-admin (backup, restore, import, memory sizing), cypher-shell (ad-hoc queries, scripting, CI/CD), or neo4j-mcp (quick MCP server install). Does NOT cover Cypher query authoring — use neo4j-cypher-skill. Does NOT cover driver upgrades — use neo4j-migration-skill. Does NOT cover full MCP editor configuration — use neo4j-mcp-skill.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

Generates, optimizes, and validates Cypher 25 queries for Neo4j 2025.x and 2026.x. Use when writing new Cypher queries, optimizing slow queries, graph pattern matching, vector or fulltext search, subqueries, or batch writes. Covers MATCH, MERGE, CREATE, WITH, RETURN, CALL, UNWIND, FOREACH, LOAD CSV, SEARCH, expressions, functions, indexes, and subqueries. Does NOT handle driver migration or API changes — use neo4j-migration-skill. Does NOT cover DB administration or server ops — use neo4j-cli-tools-skill.

日本語の概要は準備中です。原文の説明を表示しています。

neo4j-contrib/neo4j-skills1142026年10月12日 更新

neo4j-contrib のスキルをすべて見る

このスキルの問題を報告する