Memory for coding agents (MCP)

Goal: let your coding agents remember project decisions, procedures, and context between sessions.

There are two ways to do it, both built on Core LLM Wiki:

	Local: Curated Thoughts	Cloud: SynapseTree
Where memory lives	brain.db on your machine	synapsetree.dev
Status	Available now	Available now (@equationalapplications/mcp-server)
Human review gate	Yes (Review desk)	—
Harness integrations	Claude Code, OpenCode, Hermes, DeepSeek Harness, OpenClaw, Docker Agent	Any MCP client
Best for	Private, curated memory shared by a “software agency in a box”	Memory that follows you across machines and teams

Local · Curated Thoughts

1. Install Curated Thoughts for macOS, Linux, or Windows, and point it at a vault folder.

2. Install the integration for your harness from curated-thoughts-integrations. For Claude Code:

   git clone https://github.com/equationalapplications/curated-thoughts-integrations
   cd curated-thoughts-integrations/integrations/claude-code
   ./scripts/install.sh            # prints the registration; writes nothing until asked
   python3 scripts/ct_doctor.py check

3. Start a session. The integration recalls relevant curated facts at session start and on every turn (Intuitive Wisdom), and the agent can search, traverse, and deposit new wisdom through MCP tools. Nothing enters long-term memory until you approve it in the Review desk.

Tip

Any other MCP client can use the app directly: run the Curated Thoughts binary with --mcp. See its MCP Agent Server docs.

Cloud · SynapseTree

Install the SynapseTree MCP server (Node 20+):

npm install -g @equationalapplications/mcp-server

Then sign in, pick a namespace, and register it with your client:

synapsetree-mcp auth --email you@example.com
synapsetree-mcp config set namespace repo:myproject
claude mcp add synapsetree -- synapsetree-mcp

For Cursor, Windsurf, or Claude Desktop:

{ "mcpServers": { "synapsetree": { "command": "synapsetree-mcp" } } }

Check the connection:

synapsetree-mcp status
npx @modelcontextprotocol/inspector synapsetree-mcp

Tell the agent how to use memory

Whichever backend you pick, a few lines in CLAUDE.md or AGENTS.md help a lot:

## Memory
- At the start of a task, search memory for the area you're touching.
- When you make a decision that future sessions should know, record it.
- When a remembered fact is wrong, supersede it. Never delete.

The Curated Thoughts integrations ship skills native to each harness that cover this for you.