atlas_project_createLABSCO SUMMARY
Dependencies and relationships are first-class, because the store underneath is a graph.
Most task tools give you a flat list; here projects and tasks can depend on each other, knowledge attaches to a project, and one query crosses all three. atlas_deep_research is the clearest use of that: a topic goes in and comes back as a plan of linked sub-questions with tasks attached, not a document.
What it is
A project, task and knowledge manager backed by Neo4j, exposed as MCP tools. Entities carry dependencies and relationships, and one search spans all three.
What you get
- atlas_project_create, atlas_project_list, atlas_project_update and atlas_project_delete — projects with status, URLs, completion requirements and dependencies, single or bulk
- atlas_task_create, atlas_task_update, atlas_task_delete and atlas_task_list — tasks with priority, tags, assignee and dependencies, listed per project
- atlas_knowledge_add, atlas_knowledge_list and atlas_knowledge_delete — knowledge items with domain, tags and citations
- atlas_unified_search — one search across projects, tasks and knowledge, full-text or regex, with case-insensitive and fuzzy options
- atlas_deep_research — turns a topic and goal into a hierarchical research plan in the graph, optionally creating tasks for each sub-question
- Resource endpoints under atlas:// for projects, tasks and knowledge, including per-project listings
- Backup and restore to timestamped JSON exports
Requirements
A reachable Neo4j instance — the repository ships a docker-compose for one — with NEO4J_URI, NEO4J_USER and NEO4J_PASSWORD set. Node and npm to install and build. Runs over stdio by default, or as Streamable HTTP with MCP_TRANSPORT_TYPE set to http.
Setup effort
One command plus credentials — npx -y atlas-mcp-server, then supply credentials
USE CASES
01
Planning work that has to happen in order
Create tasks with dependencies so the sequence is recorded rather than remembered.
02
Keeping research findings attached to the project they belong to
Add knowledge items with domain, tags and citations, then list them per project later.
03
Finding something when you only remember a phrase
Search across projects, tasks and knowledge at once, with fuzzy matching when your memory is approximate.
04
Turning a vague question into a work plan
Give a research topic and goal and get back a hierarchical plan in the graph, with tasks created for each sub-question.
14 tools
Listed in project docs. Your client may show a different list.
atlas_project_listatlas_project_updateatlas_project_deleteatlas_task_createatlas_task_updateatlas_task_deleteatlas_task_listatlas_knowledge_addatlas_knowledge_deleteatlas_knowledge_listatlas_unified_searchatlas_deep_researchatlas_database_cleanBEFORE YOU INSTALL
What to know first
Tools you get
14 tools named in the project's documentation.
Requires
- Node.js and npm
- A running Neo4j instance — the repository's docker-compose up -d starts one, or use Neo4j AuraDB
- NEO4J_URI, NEO4J_USER and NEO4J_PASSWORD in the client configuration's env block
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
Server setup
Build from source
TERMINAL
git clone https://github.com/cyanheads/atlas-mcp-server.git
cd atlas-mcp-server
npm install
npm run buildnpx
TERMINAL
npx -y atlas-mcp-serverMCP Clients
Any MCP client
ANY MCP CLIENT · JSON
{
"mcpServers": {
"atlas-mcp-server-stdio": {
"command": "node",
"args": [
"<YOUR_PATH_TO_ATLAS_MCP_SERVER>/dist/index.js"
],
"env": {
"NEO4J_URI": "bolt://localhost:7687",
"NEO4J_USER": "neo4j",
"NEO4J_PASSWORD": "password2",
"MCP_LOG_LEVEL": "info",
"NODE_ENV": "development",
"MCP_TRANSPORT_TYPE": "stdio"
}
}
}
}<YOUR_PATH_TO_ATLAS_MCP_SERVER> is the full path of your clone — dist/index.js is created by npm run build. The npx route runs the same server from the npm package (atlas-mcp-server, bin dist/index.js) and needs no build, but still needs the NEO4J_ variables.
