Feature Request: API for claude.ai Projects file uploads (enable CLI/cron sync)

[guillermocubells]

When using Claude Code on desktop, I have full access to local project files and MCP integrations (HubSpot, etc.). But when I need to access the same context from claude.ai on mobile (browser), I have to manually upload files to a Project.

Request: Expose an API (or CLI command) to programmatically upload/sync files to claude.ai Projects. This would allow users to set up a cron job or script to keep Project Knowledge in sync with local files automatically.

Use case: As a founder, I need to query my CRM (HubSpot via MCP) with full company context while on the go from my phone. The context files rarely change, but there's no way to automate the upload — it must be done manually through the web UI every time.

This would bridge the gap between Claude Code (desktop) and claude.ai (mobile).

[MrPat420]

Use case: multi-project sync from a scripted/headless workflow

Adding a concrete case to the request above. I run several active
claude.ai Projects, each with a shared "operator brief" knowledge file
that gets revised on my own infrastructure (Linux box, version-controlled,
patch-scripted) roughly weekly. Every revision currently requires:
delete old file from each Project's knowledge panel, upload new file,
per Project, by hand, in the browser/mobile UI.

This doesn't scale past a handful of projects, and it's a bigger
problem for anyone doing this from mobile — I work primarily via SSH
from a phone, so "manually re-upload in the web UI" means a full
context-switch away from the terminal for every sync.

What would solve it: parity between claude.ai Projects and the
existing Files API (platform.claude.com/docs/build-with-claude/files).
The Files API already supports upload/list/retrieve/delete scoped to
an API key's workspace — the missing piece is a way to attach/detach
those file_ids to a specific Project's knowledge base, e.g.:

POST /v1/projects/{project_id}/knowledge_files {file_id}
DELETE /v1/projects/{project_id}/knowledge_files/{file_id}
GET /v1/projects/{project_id}/knowledge_files

Why this matters for security, not just convenience: without an
official endpoint, the only way to automate this today is unofficial
tools that authenticate with a scraped claude.ai session token rather
than a scoped API key. A session token is full account access with no
cap and no defined support contract — a real incentive exists right
now to build against an insecure, unsupported surface simply because
no sanctioned one exists. An official, API-key-scoped endpoint removes
that incentive entirely.

Happy to provide more detail on the workflow if useful.