Agent Access overview | Filepad Docs

Agent Access lets external agents operate against a Filepad workspace without becoming the main Filepad runtime. The agent gets scoped access to real files: instructions, skills, memory, sources, uploads, artifacts, and reviewed automation notes.

How it works

A workspace admin creates an Agent Access key in the workspace.
The key is granted narrow scopes such as env:read or artifacts:write.
The external agent signs every /agent-api/v1 request with HMAC-SHA256.
Filepad validates the signature, checks scopes, and serves real workspace file data.

Agent keys are workspace-scoped. A key for workspace A cannot access workspace B.

What agents can do

Read

Inspect the canonical environment folders
Read the file tree and supported text/Markdown content
Discover skills/*.md files as reusable prompts/capabilities
Search indexed workspace content
Read visible files under memory/ through the file API

Write safely

Create new note artifacts under artifacts/
Create reviewable edit proposals for supported editable files
Push activity events into Filepad

Agent Access v1 does not directly mutate .filepad/, agents/, skills/, memory/, sources/, uploads/, or automations/. Supported file edits are submitted as reviewable proposals.

Scopes

Scope	What it unlocks
`env:read`	Environment summary, file tree, supported file content, search, and prompt/resource discovery
`artifacts:write`	Create note artifacts under `artifacts/`
`files:propose`	Create reviewable edit proposals
`memory:read`	Reserved for future memory-specific read surfaces
`events.write`	Push activity events

MCP tools

Use Filepad's remote MCP endpoint when your agent supports MCP. It delegates every tool call to the backend Agent Access surface, so the same scopes, hidden-file rules, proposals, and audit trail apply.

{
  "mcpServers": {
    "filepad": {
      "transport": "streamable_http",
      "url": "https://api.filepad.ai/mcp/v1/stream"
    }
  }
}