LABSCO SUMMARY

Both sides write to the same records.

That is the design decision the rest follows from. An agent memory the human cannot edit drifts and cannot be corrected; a document the agent cannot write to goes stale. Here the notes, tasks and files are one graph with two authors, which is why the session-resume behaviour is useful rather than merely persistent. The cost is size — 184 tools is a lot of context, so this is a workspace server to run as your main one rather than one of six.

What it is

The stdio-to-HTTP bridge for CaseMgr, a shared persistent workspace for you and an AI agent. The premise is that nothing is ephemeral: the agent creates notes when it finds things, plans tasks when work appears, and marks them done; you review and edit the same records. Every session resumes where the last one stopped, and semantic search spans the whole history.

What you get

Persistent memory across sessions, so a conversation picks up from the record rather than from a summary you paste back in
A bidirectional graph both sides write to — notes, tasks, bookmarks, files, todos — rather than an agent scratchpad you cannot edit
Natural-language search across everything either of you ever wrote
A multi-agent task queue: work items dispatched, claimed and completed across several agents
Event-driven workflows as CMMN stages with sentry-gated execution, and reusable case plan models you publish, version and instantiate
File management with versioning and content-aware embeddings, workspaces and git worktrees linked across machines, and time tracking with expenses and invoices rendered to PDF
184 MCP tools in total — a large surface, and worth knowing before enabling it alongside other servers

Requirements

A CaseMgr account. Claude Desktop, Cursor and other stdio clients need this bridge and an API token in CASEMGR_API_TOKEN, from casemgr.systems/tokens; install with npm install -g casemgr-mcp (package casemgr-mcp, 1.0.1). Claude Code does not need the bridge at all — it speaks HTTP natively and authenticates over OAuth, so you point it at the server URL and skip the token. CASEMGR_URL overrides the endpoint for a self-hosted instance.

Setup effort

One command plus a key — npm install -g casemgr-mcp, then supply credentials

See installation options

USE CASES

01

Picking up where the last session ended

Open a new conversation against the same workspace and let the agent read its own notes and open tasks instead of re-explaining the project.

02

Finding something you half-remember writing

Search in natural language across notes, tasks and files, without needing to know whether you or the agent wrote it.

03

Splitting work across several agents

Put work items on the queue and let multiple agents claim and complete them, rather than running one long session.

04

Billing for the work afterwards

Use the time tracking and expenses on the case, then generate the invoice as a PDF from the same records.

05

Reusing a process you have run before

Instantiate a published case plan model instead of recreating the same task structure by hand.

BEFORE YOU INSTALL

What to know first

Requires

casemgr-mcp on PATH
CASEMGR_API_TOKEN in your environment

Where it runs

Of the 3 documented clients, 2 run the server on your machine and 1 connects to a hosted endpoint.

INSTALLATION

Connect your client

Server setup

npm

TERMINAL

npm install -g casemgr-mcp

Hosted endpoint

ENDPOINT URL

https://casemgr.systems/mcp

Add this hosted endpoint to your MCP client, then sign in.

MCP Clients

Claude Desktop, Cursor

JSON

{
  "mcpServers": {
    "casemgr": {
      "command": "casemgr-mcp",
      "env": {
        "CASEMGR_API_TOKEN": "<YOUR_CASEMGR_API_TOKEN>"
      }
    }
  }
}

Claude Code

CLAUDE CODE · JSON

{
  "mcpServers": {
    "casemgr": {
      "type": "http",
      "url": "https://casemgr.systems/mcp"
    }
  }
}

CaseMgr MCP