MCP Server for Verbatim File Storage

The agent-native file store

A private, path-addressable file store your agent writes to and reads back byte-for-byte. Save an exact artifact — a podcast transcript, a prep doc, a run-of-show — grow it over time by appending, and get the same bytes back on demand, or find it by meaning through the same semantic index that powers Brain. Files is canonical state: what you save is what you get, never a synthesized summary.

Works with Claude Code, Claude Web, OpenClaw, Claude Cowork, ChatGPT, n8n, and more.

How it works

From an exact artifact to a verbatim, recall-by-meaning file store in five steps.

1

Sign up and get an API key

Create a free account at gentic.co, generate an API key from your dashboard, and use it as a Bearer token to authenticate your agent.

2

Add the MCP server to your agent

Drop https://mcp.gentic.co/files into Claude Code, Claude Web, OpenClaw, Claude Cowork, ChatGPT, n8n, or any MCP-compatible client. Pass your Gentic API key as a Bearer token — or connect via OAuth on clients that support it.

3

Save a file

Call `save_file` with a path (e.g. `episode-2-prep/transcript.md`) and the file's content as text/markdown. The server writes the exact bytes to private storage, records the file in your org's manifest, and kicks off background indexing — both the body and the metadata (path, filename, tags, title) become searchable. Free.

4

Get it back, exactly

`get_file` returns the precise UTF-8 bytes you stored — no summarization, no truncation. Ownership is verified against your org's manifest before any read, so files stay private to the org that saved them. `list_files` walks a folder by path prefix when you want to see what's there.

5

Find by meaning

`find_file` searches the same semantic index that powers Brain — filtered to your files — and joins the manifest so you get back the matching paths to fetch. It's retrieval-only: it finds which file you mean, then `get_file` returns it verbatim. Scope a search to a folder by passing a path prefix.

Available tools

6 tools your AI agent can call through the Model Context Protocol.

append_file

Free

Append content to the END of a file without resending the whole thing — the server reads the existing bytes and concatenates yours after a separator (default newline), so the prior content is preserved exactly.

delete_file

Free

Delete a file for real — removes the stored object, tombstones the manifest record, and unindexes it.

find_file

Free

Find your saved files by meaning when you don't remember the exact path — e.g.

get_file

Free

Retrieve a file's exact, verbatim content by its path.

list_files

Free

List your saved files, optionally scoped to a folder via `prefix` (folders are just path prefixes).

save_file

Free

Save a file VERBATIM to your private, org-owned file store and register it in the files manifest.

Connect in seconds

Sign up, grab your API key, and add the MCP server to your agent. Works with Claude Code, Claude Web, OpenClaw, Claude Cowork, ChatGPT, n8n, and any MCP client.

claude mcp add gentic-files \
  --transport http \
  https://mcp.gentic.co/files \
  --header "Authorization: Bearer YOUR_API_KEY"
OAuth
Using Claude or ChatGPT websites?

You can also connect via OAuth — just add gentic-files as a connector in Claude or ChatGPT settings. No API key needed; authentication is handled automatically.

Recommended
Supercharge with Agent Skills

Install the Gentic agent skill to teach your agent the optimal workflow so it gets the best results automatically.

npx skills add gentic-co/agent-skills

Works with Claude Code, Cursor, Copilot, and 40+ other agents.

Transparent per-call pricing

Every Files tool is free. Background indexing charges its own per-chunk meter on the shared vectorize_content index — you only pay for the indexing you use.

ToolCost
append_fileFree
delete_fileFree
find_fileFree
get_fileFree
list_filesFree
save_fileFree

Pricing is pulled live from the Gentic MCP manifest so it always matches what you'll actually be charged.

Why agent-native?

Cloud drives assume a human dragging files into folders. Files assumes an agent that saves an exact artifact, recalls it by meaning, and hands the user back the real thing — byte-for-byte.

Verbatim by contract

What `save_file` stores is what `get_file` returns — the exact UTF-8 bytes, no synthesis or truncation. A saved file is canonical state, never a derived summary. Save a transcript, get that transcript back word-for-word.

Private by default

Files are written to a private store — never a public URL. Every read verifies per-org ownership against the manifest before touching storage, so one org can never read another's files. Org-scoped isolation is enforced on the read boundary, not by obscurity.

Recall by meaning

Saving a file indexes both its body and its metadata into the same knowledge layer that powers Brain. `find_file` does retrieval-only semantic search over just your files — so 'the transcript from episode 34' resolves even when you don't remember the path.

Free to use

Every Files tool is free — save, get, list, find, and delete. Background indexing charges its own per-chunk meter on the shared `vectorize_content` index when a saved file is vectorized, the same note as Brain. You only pay for the indexing you use.

Give your agent a verbatim file store

Sign up, grab your API key, connect the MCP server, and let your agent save exact files, recall them by meaning, and hand you back the real thing.