Thin MCP stdio client for ModelBrain, a local memory layer for AI assistants that runs on your own
Thin MCP stdio client for ModelBrain, a local memory layer for AI assistants that runs on your own computer.
·
A thin MCP stdio client for ModelBrain: the private memory layer for AI assistants: one memory for every assistant you use, kept on your own computer.
This package does not install or bundle ModelBrain. It connects to the ModelBrain desktop app already running on your machine.
The launcher still starts and answers the MCP handshake, so your client shows the server and its tools instead of a start-up failure. Every tool call then returns one plain result: ModelBrain is not installed, and where to get it. Nothing is stored or recalled in this state. The same message is printed to standard error.
The tool list it shows comes from bin/tools.json, a copy of what the real server reports. Maintainers refresh it with npm run snapshot-tools on a machine that has ModelBrain installed.
To get the old behaviour, where the launcher exits with status 1 when the app is missing, pass --require-app or set MODELBRAIN_MCP_REQUIRE_APP=1.
ModelBrain's own installed app already ships a binary, mcp_server, that speaks MCP over stdio and talks to the local ModelBrain daemon. This package finds that binary on the current machine and runs it. It exists so ModelBrain can be registered with the MCP Registry and the directories that list from it, with a reviewable package as the distribution target, while the actual engine stays out of this repository.
In ModelBrain's Allow or Block prompt this client shows up as your assistant's name with node as the program (mcp_server.exe on Windows). If you turn on Require client tokens in ModelBrain, mint a token for your assistant with modelbrain_admin setup tokens --mint NAME and put it in your client's env block as MODELBRAIN_CLIENT_TOKEN; the launcher passes it through.
Most MCP clients run a server with npx. Point your client's MCP config at:
{
"mcpServers": {
"modelbrain": {
"command": "npx",
"args": ["-y", "modelbrain-mcp"]
}
}
}Claude Desktop, Claude Code and Cursor users do not need this: ModelBrain's own app registers itself with those clients directly. See modelbrain.net/docs for the full tool surface (14 tools: remember, recall, list_memories, expand, forget, six partition tools, diagnostics, usage, ping) and the connect guides for other clients.
MIT, for this wrapper package. ModelBrain itself is separate, proprietary software; see modelbrain.net for its own terms.
See SECURITY.md.
Pick your client and paste the snippet. Each one is the same server, written the way that client expects it.
claude mcp add modelbrain -- npx -y modelbrain-mcp{
"mcpServers": {
"modelbrain": {
"args": [
"-y",
"modelbrain-mcp"
],
"command": "npx"
}
}
}code --add-mcp '{"name":"modelbrain","args":["-y","modelbrain-mcp"],"command":"npx"}'[mcp_servers.modelbrain]
command = "npx"
args = ["-y", "modelbrain-mcp"]Runs locally on your device. Your client starts the server itself, so nothing has to be hosted.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/modelbrain-mcp to learn what the "Modelbrain Mcp" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
| Name | What it does |
|---|---|
ping | Liveness check that never touches the store. Use `diagnostics` for anything more. |
recall | Find the memories nearest a query, best first. Required: `query`. Check this before saying you don't know something about the user; it is not consulted automatically. Hits carry `key`, `text`, `releva |
remember | Save a fact, preference or decision so a later conversation can recall it. Use it when the user asks you to remember something, and whenever the user states a fact, preference or decision worth keepin |
expand | Fetch one memory by key: full text, tags and provenance. A split memory comes back whole whichever piece you ask for (`chunk_keys` lists the pieces; `piece_only: true` returns just the one). `truncate |
forget | Forget a remembered object by key. `mode` defaults to "revert": on a chain head created with `supersedes` it tombstones the head and restores the predecessor; otherwise it tombstones the target. "chai |
list_memories | List memories by scope without a query, newest first. Use `scope: "user"` for what the user told you, or `scope: "connector:<id>"` for a synced source. Filters work as in `recall`. `count` is the tota |
partition_list | List partitions, the namespaces that keep memories apart (for example one per client). Every store has a default "unsorted" partition. `include_archived: true` also lists archived ones. |
partition_create | Create a partition by name. The name may not be "current", "all", "all_active", "none", start with "id:", or match an active partition. Call `partition_use` to save into it. |
partition_use | Bind this connection to one partition (an unbound connection searches every active partition): later `remember` calls without `partition` land there, and `recall`/`list_memories` without `partition_sc |
partition_rename | Rename a partition by id. Refused for the default partition and for reserved or taken names. |
partition_archive | Archive a partition by id. Its memories stay reachable with `partition_scope: "id:<ulid>"`. Refused for the default partition. Idempotent. |
partition_restore | Restore an archived partition by id. Refused if an active partition now has its name. Idempotent. |
diagnostics | Report health: version, store location, `live_count`, search mode, uptime, connector sync state and degraded status. `live_breakdown` explains why `list_memories` may show fewer than `live_count`. Che |
usage | Summarize this server's local tool-call log over the last `days`: calls per session by tool, empty recalls, saves per session and errors by code. The log holds no memory content. Not needed for normal |
Read live from the server on November 12, 2026.
Paste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/modelbrain-mcp)Official FlowchartAI remote MCP connection configuration, OAuth setup, and plugin manifests.
Read-only germline variant evidence, gene profiles and scientific literature via MCP
Shared skills, prompts, and tooling for Picsart's gen-ai CLI and generative-AI workflows
No tagged releases on record.
Connect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "modelbrain-mcp" with a rating from 1 to 5 and a short note on what worked and what did not.No reviews yet
Be the first to share your experience with this tool.
Official FlowchartAI remote MCP connection configuration, OAuth setup, and plugin manifests.
Developer Tools
Read-only germline variant evidence, gene profiles and scientific literature via MCP
Developer Tools
Shared skills, prompts, and tooling for Picsart's gen-ai CLI and generative-AI workflows
Developer Tools
One email a week. New agents, MCP servers and skills, and what is actually getting traction.