Binder
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Binderadd a record titled 'Dark mode' to my Batches list"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Clavarium Binder
Lists of records, kept together by people and AI agents, each with its id.
A binder holds lists: batches of work, development tasks, customers, people, a roadmap, anything followed over time. A list starts from a template (its states, its fields, the view it opens in) and everything is changed afterwards. A record has a title, a description, a state, its fields, short notes, as many things attached as wanted (messages of a chat, files, assets, branches, links, other records), owners, a due date (with an optional lead reminder a few days before it), and its history.
binder your lists
binder show Batches the records of a list, by state
binder add Batches "Dark mode" --description "The whole site in dark, for evening reading."
binder note 12 "the login page only"
binder attach 12 branch site:dark-mode
binder move 12 Doing past a gate, only once a person approved itThe description is for a person, read at a glance: short and clear, what it is
and why, in 1 to 3 lines (paragraphs, - lists, **bold** and links are
shown as such; 4000 characters at most). An agent writes one for each record it
adds.
An MCP server gives the same verbs to an AI agent. A person sees the binder on
its screen (ui/), drawn by Binder in the page of the application that runs it.
A sibling of Clavarium Parley and Clavarium Docket, built the same way, universal (nothing in it is specific to Clavarium).
What an agent cannot do
Rewrite the binder. It is held by its keeper (the application, or
binder serve), out of the agent's reach. An agent asks through its own folder; the keeper answers, and says who asked by that folder.Approve. A list may have a gate (Batches: "Approved"): a record reaches it, or any state after it, only once a person ticked it. An agent proposes, and works on what was approved.
See what it was not given. Access is given per list: read, or change. It may make a list of its own from a template (
binder new batches): given to it alone, and seen by a person as any other; a few at most.Delete. It archives; a person deletes, and deleted is deleted.
Related MCP server: last-mile-mcp
Without an application
binder serve --store ~/binder-data --side ./.binder=Assistant --lang frkeeps a binder and shows its page on http://127.0.0.1:4180/. Put the store
where the agents cannot write.
See docs/protocol.md. No dependency; Node 22 or later.
Status: first version (0.1), not yet published.
License: MIT.
This server cannot be deployed
Maintenance
Related MCP Connectors
Context and control for any AI. Agent identity, scoped access, and proof across tools.
Governed memory and workspace for any AI: tasks, calendar, mail and pages, with per-action consent.
Hosted CRM and email desk for AI agents: key per agent, audit log, one-click undo, opt-in approval.
The system of record for AI agent authority: playbooks, routed policy questions, reusable rules.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceGoverned, self-hosted memory for AI agents: writes queue until an authorized approver signs off.Apache 2.0
- AlicenseBqualityCmaintenanceEnables AI agents to act on live business objects under enforceable per-call identity, per-tool grants, and mandatory human approval for irreversible actions, with connectors isolated from core logic.8MIT
- AlicenseBqualityBmaintenanceEnables AI agents to execute actions under an accountability layer with identity passports, mandates, a permission gate, and a tamper-evident journal, while requiring human confirmation for irreversible operations.7125 PyPI5AGPL 3.0
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to list entries and metadata, run commands with secrets injected as environment variables while scrubbing every value from the output, copy secrets straight to the clipboard, and generate or store new secrets without their plaintext ever entering the conversation. Obvious dumpers are refused and destructive edits require explicit confirmation, closing off accidental exposure paths.71 PyPIMIT