casemgr-mcp
# casemgr-mcp
[](https://smithery.ai/servers/mstang/casemgr)
**A shared workspace your AI agent actually writes to — 184 MCP tools.**
It creates notes when it finds things, plans tasks when work appears, and marks them done. You review and edit. Persistent memory + semantic search across a graph of notes, tasks, and files.
This package is the stdio-to-HTTP bridge for the [CaseMgr](https://casemgr.systems) MCP server — use it with any MCP client that supports stdio transport (Claude Desktop, Cursor, etc.). For Claude Code, see [below](#claude-code-cli) — it supports HTTP natively and does not need this bridge.
## Quick Start
```bash
npm install -g casemgr-mcp
```
Get an API token at [casemgr.systems/tokens](https://casemgr.systems/tokens).
## MCP Client Config
### Claude Desktop / Cursor
Add to your MCP settings:
```json
{
"mcpServers": {
"casemgr": {
"command": "casemgr-mcp",
"env": {
"CASEMGR_API_TOKEN": "your_token_here"
}
}
}
}
```
### Claude Code (CLI)
Claude Code authenticates via OAuth automatically — no token needed. Just add the server URL:
```json
{
"mcpServers": {
"casemgr": {
"url": "https://casemgr.systems/mcp"
}
}
}
```
If you prefer token auth, add `"headers": { "Authorization": "Bearer your_token_here" }`.
### Smithery
Install directly via [Smithery](https://smithery.ai/servers/mstang/casemgr) — one-click setup for Claude Code, Cursor, Codex, and more.
## What is CaseMgr?
A shared, persistent workspace for you and your AI agent — 184 MCP tools across a graph of notes, tasks, files, calendar, and agent presence.
Your agent creates notes when it finds things. Plans tasks when there's work to do. Marks them done as it finishes. You review, edit, and add your own. Nothing is ephemeral — every session resumes exactly where the last one left off, and semantic search spans everything either of you ever wrote.
**What's in the workspace:**
- **Persistent memory** — every session resumes where the last one left off; nothing is ephemeral
- **Bidirectional graph** — agent and user both read and write notes, tasks, bookmarks, files, todos
- **Semantic search** — natural-language search across everything either of you ever wrote
- **Multi-agent task queue** — dispatch, claim, and complete work items across multiple agents
- **Event-driven workflows** — CMMN stages with sentry-gated execution and auto-lifecycle management
- **Reusable templates** — case plan models with publish, version, and instantiate
- **File management** — upload, download, versioning, content-aware embeddings
- **Workspaces & worktrees** — organize cases, link git worktrees across machines
- **Time tracking & billing** — durations, expenses, invoices with LaTeX/PDF generation
## Environment Variables
| Variable | Required | Default | Description |
|----------|----------|---------|-------------|
| `CASEMGR_API_TOKEN` | Yes (stdio) | — | API token from casemgr.systems/tokens. Required for stdio clients (Claude Desktop, Cursor). Not needed for Claude Code, which uses OAuth. |
| `CASEMGR_URL` | No | `https://casemgr.systems/mcp` | MCP endpoint URL (for self-hosted instances) |
## License
MIT
TDQS
Scored across 205 tools
Namespace prefixes help distinguish tools, but overlapping concepts across namespaces — e.g., 'task'/'todo' appears in items-, cmmn-, and tasks- with different semantics, and multiple sentry-evaluation tools (cmmn-check_sentry_condition, cmmn-evaluate_sentry, cmmn-mark_sentry_satisfied) require careful reading. Detailed descriptions mitigate this, but the volume of 205 tools makes selection error-prone.
Tool names consistently follow a `namespace-verb_noun` pattern with snake_case, which is highly predictable. Minor deviations exist: `meta-desire_paths` and `meta-tool_stats` are noun phrases, `calendar-get_or_create` uses a compound verb, and `models-delete_definition` is an alias of `models-remove_definition`. These are small exceptions in a largely uniform convention.
With 205 tools across 20+ namespaces, this is an extremely large surface for any MCP server. Even though the domain is broad, the sheer number overwhelms agents and violates the typical well-scoped server expectation. Many tools could be consolidated or exposed lazily.
The tool set provides CRUD and lifecycle coverage for nearly all entities in the domain: cases, items, models, clients, invoices, users, workspaces, worktrees, and more. Minor gaps exist — e.g., no direct message delete (only archive), no calendar event management, and documentation entries lack an update operation — but agents can work around these.