fileaify.Augment any AI agent
Remote MCP server

Fileaify MCP connector

A hosted MCP server that gives your agent your project's documentation, decisions and controlled access.

Connect it

The server address is the same for every client:

https://fileaify.com/mcp
  1. Create a free workspace and a project.
  2. Add a custom connector in your MCP client, using that address. Claude and ChatGPT are supported through published client identity documents, and other clients can use dynamic registration. Availability depends on your client and its administrative settings.
  3. Approve the connection in your browser. You choose one workspace, its projects, the permissions and a duration of 1 to 30 days.

There is no client secret to paste into a chat and nothing to install. The transport is Streamable HTTP with OAuth and mandatory PKCE.

No MCP client? Your agent can also read fileaify.com/agent and use the same operations over REST.

What your agent can do

  • Read the project: an index, then only the documents and sections the task needs, plus their history.
  • Know the rules: the project's contract says where each piece of information belongs and how to propose a change.
  • Propose changes: create and edit documents through reviewable proposals, with version checks. Your publication policy decides what goes live.
  • Find services and request access: see what the project uses, and request a precise permission with a duration. A person decides.
  • Run work: where the plan and grants allow it, run commands in the executor, with keys used on the server side.

What the connection does not grant

The consent screen grants no secret value and no permission to use a secret, and an agent cannot approve its own requests. Those are separate decisions, recorded and revocable.

  • Access tokens last at most one hour, and refresh is limited by the expiry you approved.
  • The connector appears as an agent in Fileaify, where you can suspend or revoke it.
  • Delivering a secret's value to a client is disabled by default. If an administrator enables it and a person approves it, the value does reach the client and may enter the model's context, so prefer use on the server side when the task allows it.

Details

The full tool list, limits and OAuth endpoints are in the MCP reference. No internal AI is involved, and using the connector consumes no assistant quota.