Persistent memory for AI agents

LLM context windows end. Sessions expire. The research your assistant did on Monday is gone by Tuesday, and each new chat re-explains the same preferences, links, and decisions. Most "memory" features are locked inside one vendor's product: what Claude remembers, Cursor can't see, and you can't browse, edit, or export any of it.

Turbodoc is a knowledge base with a hosted MCP server. Any MCP-compatible agent — Claude, Claude Code, Cursor, VS Code, Windsurf — gets 22 tools to create, search, update, and organize four content types:

• Bookmarks — every source an agent finds, with tags, read status, and automatic metadata.
• Notes — markdown summaries, decisions, and long-term context an agent should recall later.
• Code snippets — reusable functions, configs, and commands with syntax highlighting.
• Diagrams — Mermaid diagrams an agent can generate to document an architecture or flow.

Because the store lives outside any single AI product, it's shared memory: your coding agent saves a snippet, your chat assistant recalls it next week, and you can read and edit all of it in the Turbodoc apps on web and iOS.

During a research session:

"Search for recent papers on RAG evaluation, bookmark the
best five in Turbodoc tagged 'rag-eval', and write a note
summarizing the main approaches."

A week later, in a completely new conversation — or a different tool:

"Check my Turbodoc notes and bookmarks tagged 'rag-eval'
and draft an evaluation plan based on what we found."

The agent calls search_bookmarks and list_notes, retrieves exactly what was saved, and continues where the last session stopped.

• Human-readable by design. Memory an agent writes is a library you actually use — browsable, searchable, and editable in a real app, not rows in a table.
• Zero infrastructure. No database to host, no embeddings pipeline to maintain. Connect the MCP endpoint and you're done.
• OAuth-scoped security. Agents authenticate as you, see only your data, and access is revocable at any time.
• Open source. The entire stack is public at github.com/turbodoc-org — no lock-in on the thing holding your memory.

Create a free account, then follow the MCP setup guide for your client. For Claude Code it's one line:

claude mcp add --transport http turbodoc https://api.turbodoc.ai/mcp