Persistent memory
for your AI agents.
Connect Claude Code, Cursor, or any MCP-compatible agent to your Nilon vault. Agents save decisions, recall past context, and share a live memory layer across sessions and projects.
Cross-session memory
Agents write session summaries, architecture decisions, and outputs to the vault. Every future session reads from the same store — context accumulates rather than resetting.
Semantic-ready search
search_vault finds relevant past saves before your agent starts new work. No more re-solving problems that were solved two months ago in a different project.
Human-readable
Everything agents write appears in your Nilon vault UI. Browse, annotate, share, and export agent-written saves alongside your own — full visibility into what your agents know.
Multi-agent pipelines
One vault.
Multiple agents.
All API keys from the same Nilon account share the same vault. Multiple Claude Code instances — or agents in a pipeline — can read and write simultaneously.
- 1Agent A saves an architecture decision via save_to_vault
- 2Agent B calls search_vault before starting related work
- 3Agent B extends the same document via update_save
- 4Agent C adds a note with observations via add_note
Six tools
Everything your agent needs
save_to_vaultPersist any string to your vault — decisions, code, research, or outputs worth keeping. Auto-tagging runs asynchronously after save.
save_summaryPaste key exchanges from the current session. Claude Haiku condenses them into a structured summary (What was done / Key decisions / Next steps) and saves it.
search_vaultKeyword search across all saved content, titles, and notes. Returns matching saves with full content and metadata. The core recall tool.
list_recentReturn the N most recent saves. Call at the start of a session to establish context from recent work without needing a specific query.
update_saveReplace the content of an existing save. Tags, title, and notes are preserved. Use to keep a living document current rather than accumulating duplicates.
add_noteAttach a short annotation to an existing save without changing its content. Agents leave breadcrumbs for future sessions or for other agents in the same pipeline.
Setup
Add Nilon to Claude Code in 2 minutes
- 1
Generate an API key
Go to Settings → Developer and click New API key. Copy the key — it is shown once.
- 2
Create .mcp.json in your project root
Paste the config below, replacing YOUR_API_KEY with the key you just copied.
- 3
Add .mcp.json to .gitignore
The file contains your API key. Never commit it to source control.
{ "mcpServers": { "nilon": { "type": "http", "url": "https://mcp.nilon.io/mcp", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } }}
Why not just use CLAUDE.md?
CLAUDE.md
- Static — must be manually updated
- Loads entire file every session
- Scoped to a single repository
- No search — agent reads the whole thing
- Invisible to you until you open the file
Nilon MCP
- Dynamic — agents write new entries automatically
- search_vault finds the relevant save, not everything
- Global to your account — works across all projects
- Agents can update and annotate existing saves
- Full visibility in the Nilon vault UI
Common questions
Can multiple agents share the same vault?
Yes. All API keys from the same Nilon account access the same vault. Multiple Claude Code instances or a multi-agent pipeline can read and write simultaneously — every agent sees the same shared memory.
Does this work with agents other than Claude Code?
Any MCP-compatible client works with the Nilon server. The server uses the standard Streamable HTTP transport from the MCP SDK.
Are there rate limits?
Writes (save, update) are limited to 100 per hour. Reads (search, list) are limited to 200 per hour per API key. Free plan users have a 50-save vault limit.
Is the type: "http" field required?
Yes — Claude Code requires the type field to recognise HTTP MCP servers. Without it the server will not appear in the tool list.
Your agents deserve better than a static file.
Connect your agents to Nilon in 2 minutes. Free to start.