eink-mcp
Provides tools for storing and managing files in Vercel Blob storage, including uploading content or files from URLs, retrieving, deleting, and listing files with immutable public URLs and a sync manifest for e-ink devices.
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., "@eink-mcp publish the new app.js bundle to the kindle"
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.
eink-mcp
An MCP server that hosts files for e-ink devices and syncs them down.
Agents put files here with MCP tools. Each file version lands once at an immutable,
content-addressed public URL in a Vercel Blob store, and a small manifest.json maps paths to
those URLs with their sha256. Only the manifest is ever overwritten, so a reader can never see a
manifest that disagrees with the file it names; it can only be up to a minute behind. A device such as a
jailbroken Kindle running eink-ui points its update URL at
that manifest and pulls whatever changed on every start, wake and reload. Nothing has to reach
the device: it is behind NAT and asleep most of the day, so it always pulls.
agent ──MCP──▶ eink-mcp ──put──▶ Vercel Blob (public URLs + manifest.json) ◀──pull── deviceTools
tool | effect |
| the public base URL to set as the device's update URL, the manifest URL, file count |
| every file with size, sha256, URL and last update |
| upload |
| download a URL into the store, for large builds coming out of CI |
| read a file back (text or base64, truncated past 512 KiB) |
| remove a file; devices drop their copy on the next sync |
Paths are relative and map straight onto the device's app directory. On eink-ui devices app.js
is the React bundle and eink-host is the host binary; changing either restarts the host.
Related MCP server: slm-mesh
Run it
Environment:
BLOB_READ_WRITE_TOKEN(required): a Vercel Blob read-write token. This is the only credential; the store is the only state.EINK_PREFIX(defaulteink/): the folder inside the store, one per device or fleet.
npx eink-mcp # from npm, stdio
nix run github:sberan/eink-mcp # from the flake, stdioClaude Code
claude mcp add eink -e BLOB_READ_WRITE_TOKEN=... -- npx eink-mcppg_mcp (Nix runtime)
pg_mcp builds the flake at a pinned commit and runs it over stdio; the Blob token is supplied
as the server's bearer credential through token_env.
SELECT mcp.register_server('eink', auth => 'bearer',
options => '{"token_env":"BLOB_READ_WRITE_TOKEN","env":{"EINK_PREFIX":"kindle/"}}'::jsonb,
runtime => '{"kind":"nix","source":"github:sberan/eink-mcp","revision":"<commit>",
"attribute":"mcp","executable":"bin/eink-mcp"}'::jsonb);
-- then set the bearer token (the Blob token) and wait for mcp.status('eink').runtime = readyManifest
{ "version": 1, "generated": "2026-09-17T01:59:51Z",
"base": "https://<store>.public.blob.vercel-storage.com/kindle/",
"files": { "app.js": { "sha256": "…", "size": 123304, "url": "…", "content_type": "application/javascript", "updated": "…" } } }A device keeps the manifest's ETag and the sha256 of each file it holds, downloads only files whose hash changed, verifies the hash (the CDN can lag a file behind its manifest for up to a minute, in which case the device simply tries again on its next sync), and deletes files that left the manifest.
Develop
npm install && npm run build
BLOB_READ_WRITE_TOKEN=... npm run smoke # end to end against a throwaway prefixAvailable Tools
6 toolsdelete_fileDelete a fileA
Remove a file from the store and the manifest. Devices delete their copy on the next sync.
Inputs: path (required).
get_fileGet a fileA
Read a file back: text for text types, base64 otherwise. Large files are truncated; use the URL from list_files for the whole thing.
Inputs: path (required).
list_filesList filesA
Every file in the store with its size, sha256, public URL and last update.
Inputs: no named parameters.
put_filePut a fileA
Upload a file by path (relative, as it will appear in the device app directory). Give text as content or binary as content_base64. Devices pick it up on their next sync.
Inputs: path (required), content, content_type, content_base64.
put_file_from_urlPut a file from a URLA
Download a URL and store it under path, so large binaries (a host build from CI, for example) never pass through the conversation.
Inputs: url (required), path (required), content_type.
statusStore statusC
The public base URL devices sync from (set it as update_url on the device), the manifest, and the file count.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool targets a distinct operation: status reports server-level info, list_files enumerates metadata, put_file and put_file_from_url are clearly separated by input source, get_file retrieves content, and delete_file removes files. There is no meaningful overlap that would cause an agent to select the wrong tool.
Most tools follow a consistent verb_noun pattern: list_files, put_file, get_file, delete_file, and put_file_from_url extends that pattern naturally. The single outlier is status, which is a noun rather than verb_noun, but it is a conventional command name and not confusing.
Six tools is well-scoped for a file store sync server. Each tool covers a necessary operation without redundancy, and the count is neither too thin nor too heavy for the stated purpose.
The tool surface covers the full file lifecycle: create/update via put_file and put_file_from_url, read via get_file and list_files, delete via delete_file, plus status for device sync configuration. There are no obvious missing operations for the domain.
Related MCP Connectors
Publish HTML, files, or a URL to a permanent public URL, then update it — from any MCP agent.
Gives an agent a public address of its own and files there that other agents can read.
Publish, update, read, rename, and share single-URL web pages from any AI agent.
Publish websites with a small backend (KV, SQLite, files). MCP for AI agents.
Related MCP Servers
- AlicenseAqualityCmaintenanceSimple and free publishing of content on the web for AI Agents29,324 npmMIT
- AlicenseNot gradedqualityDmaintenanceEnables peer-to-peer communication, discovery, shared state, and file coordination between AI coding agents across machines and sessions.8 npm16Elastic 2.0

A2AL MCP Serverofficial
AlicenseNot gradedqualityBmaintenanceEnables AI agents to publish themselves, discover each other, and establish authenticated encrypted connections without central infrastructure, using a decentralized agent-to-agent networking protocol.1Mozilla Public 2.0
openfuse-mcpofficial
AlicenseNot gradedqualityDmaintenanceEnables AI agents to manage persistent context, send signed messages, sync with peers, and share files through plain-file-based tools.54 npm18MIT