manifest-mcp
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., "@manifest-mcpshow me my manifest"
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.
manifest-mcp
The server behind the link option at readthemanifest.net — one personal connector URL that lets Claude and ChatGPT read your manifest (your portable context) and save new facts to it.
This repo exists so you don't have to take our word for what the server does. Read it. It's three small files.
Prefer to hold the files yourself, with no server at all? The same context layer as a plain repo you own is Context OS — that is the "Do it yourself" lane on the site.
How it treats your manifest
Your connector URL looks like https://readthemanifest.net/mcp/<token>. The token is a
long random secret, and it does two jobs:
It's the address. The server hashes the token (SHA-256) and uses the hash to label your storage slot. The token itself is never written down — not in a database, not in a log we keep.
It's the key. Your manifest is encrypted (AES-256-GCM) with a key derived from the token (HKDF). So what sits in storage is a locked file the server cannot open on its own. It is only unlocked in the moment your AI asks for it, because the request carries the key.
No accounts, no email, no passwords. The honest consequence: lose the URL and the manifest is unrecoverable — there is nothing to reset. Export a copy anytime.
Related MCP server: CacheTank MCP Server
What "open source" does and doesn't prove
Reading this code tells you what the published server does. It cannot prove what a remote server is running — no open-source project's deployment can. If that distinction matters to you, Manifest's GitHub options keep your context entirely in your own hands, and this repo lets you run your own copy of this server (below) so the deployment is yours too.
The files
netlify/functions/mcp.mjs— the MCP endpoint (POST /mcp/:token). Stateless JSON-RPC over HTTP. Two tools:get_manifest(read) andadd_to_manifest(append one dated fact).netlify/functions/manifest-vault.mjs— vault management (POST /api/manifest-vault): create, replace, export, delete.netlify/functions/_lib/vault.mjs— the storage and crypto described above. Storage is Netlify Blobs.
Limits: manifest ≤ 150KB, note ≤ 4KB.
Run your own
Deploy this repo to any Netlify account (free tier works; Blobs is enabled automatically),
then add your own https://<your-site>/mcp/<token> as a custom connector. To try it
locally:
npm install
npx netlify dev --offline
# create a vault
curl -s -X POST http://localhost:8888/api/manifest-vault \
-H 'Content-Type: application/json' \
-d '{"action":"create","manifest":"# Manifest\n\nHello."}'
# read it the way your AI would (use the returned token)
curl -s -X POST http://localhost:8888/mcp/<token> \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_manifest","arguments":{}}}'(Offline local dev stores vaults as files under .netlify/local-blobs/; deployed, it's
Netlify Blobs with strong read-after-write consistency.)
Usage counting
The server tallies how often its endpoints get used — an anonymous counter per
event name (vault_create, mcp_read, …) plus a per-day variant, in a separate
blob store. That's the whole record: no IPs, no user agents, no tokens, no
manifest content. See netlify/functions/_lib/counters.mjs — it's ~30 lines.
License
MIT.
This server cannot be deployed
Maintenance
Related MCP Connectors
Governed memory and workspace for any AI: tasks, calendar, mail and pages, with per-action consent.
Person-owned AI memory that learns, not just stores — portable context for any MCP client.
- OpenOakOAuthorg.openoak
Secure AI access to OpenOak tasks, notes, and Kanban boards.
- JustOnceOAuthai.justonce
Persistent memory for AI assistants — one shared, OAuth-secured vault for every MCP client.
Related MCP Servers
- AlicenseDqualityDmaintenanceEnables AI assistants to access and update personalized context data, creating persistent memory between sessions.14MIT
- AlicenseAqualityFmaintenanceProvides persistent personal context (identity, projects, decisions, knowledge) to MCP-compatible AI tools, eliminating the need to re-explain yourself across sessions.251 npm1MIT
- FlicenseAqualityDmaintenanceProvides persistent personal context storage across AI conversations, allowing AI assistants to remember user preferences, project conventions, and other personal information between sessions.8-
- FlicenseCqualityDmaintenanceProvides structured personal data (skills, experiences, values) as context to AI services, enabling personalized outputs across different AI applications.100-