Protocol
MCP Server for Wenlan in Claude Code, Codex, ChatGPT, and Cursor
Learn what an MCP memory server does, how it connects AI tools to durable context, and how Wenlan keeps that memory local and inspectable.
Article packet
Concepts
Developers and AI power users connecting multiple MCP clients
7 min read
01
MCP servers expose tools, resources, and prompts to AI applications through a standard protocol.
02
A memory server gives clients a way to store, search, recall, and manage durable work context.
03
Wenlan uses local MCP for coding tools and Streamable HTTP MCP for ChatGPT and Claude.ai, with the same source-backed wiki behind both paths.
01
Quick answer
If you searched for an MCP memory server, you probably want an AI tool to remember project facts, decisions, preferences, and handoffs across sessions without pasting the same context every time.
Wenlan is the local-first version of that workflow: the daemon owns the memory store, wenlan-mcp is the connector clients launch, and MCP-compatible tools call Wenlan when they need context.
- Use MCP when the memory should be available outside one chat product.
- Use a local daemon when the memory includes private project context.
- Use Wenlan when you also want capture, recall, handoff, provenance, wiki pages, and human-readable artifacts.
02
What MCP adds to memory
The Model Context Protocol gives AI clients a standard way to call external capabilities. MCP servers expose those capabilities; MCP clients are created by host applications such as IDEs or AI tools to talk to particular servers.
For memory, that protocol boundary matters. The AI tool can stay focused on the conversation while the memory server handles durable context, search, and storage.
03
What a memory server should expose
A useful MCP memory server should do more than append notes. It needs a way to capture one durable fact, recall relevant context, list or inspect stored memory, forget stale entries, and diagnose whether the local route is connected.
Wenlan exposes that path through MCP tools around context, capture, recall, distillation, review, forget, and doctor checks. The goal is not a raw database endpoint; the goal is a memory loop agents can use safely during work.
04
Local vs hosted memory
Hosted memory servers are easy to start, but they require sending memory to someone else's infrastructure. Local memory servers take more care, but they keep private project context, preferences, and decisions under your control.
Wenlan is built around the local-first path. The daemon runs on your machine, owns the database, and serves memory to MCP clients through the local wenlan-mcp connector.
05
Install path
Claude Code and Codex have plugin paths. Local MCP clients should run Wenlan setup, then use wenlan connect <client> to write the client-specific configuration.
ChatGPT and Claude.ai use Streamable HTTP MCP. The guided path is the desktop app's Remote Access panel, which creates the URL and shows each web client's setup steps.
MCP client setup
npx -y wenlan setup
~/.wenlan/bin/wenlan connect cursor
~/.wenlan/bin/wenlan connect codex
# or: claude-desktop, vscode, gemini06
How Wenlan fits
Wenlan is more than a bare MCP store. It is a source-backed LLM wiki with a local runtime, CLI, MCP connector, Claude Code plugin, Codex plugin, optional desktop app, and human review paths.
The MCP server is the bridge: AI tools read and write memory, while Wenlan keeps the broader work context visible, searchable, and locally owned.
Connect an MCP memory server locally
Install Wenlan, add the MCP connector to your client, and verify capture and recall before trusting memory in real work.
FAQ