Cody forgets your project the moment the chat ends
MCP fixes that, on Enterprise.
Your decisions. Your stack. Where you left off. Bridge MemoryRouter into Cody's agentic context fetching and pull it all back.
Five minutes. Permanently.
Before you start: Cody's MCP support runs through agentic context fetching, which requires Sourcegraph Enterprise with the agentic-context-mcp-enabled feature flag turned on by an admin. It currently supports local (stdio) MCP servers only, not direct remote HTTP connections. Because MemoryRouter is a remote, OAuth-authenticated server, this guide bridges the two with mcp-remote.
14 days free
Full access. Bridge it in, ask Cody to remember, cancel anytime.
See the difference
You: Continue the search index change we planned
Cody: I do not see a prior plan in this chat. What index and what change?
A new chat, and yesterday's plan is gone.
You: Check memory, then continue the search index change we planned
Cody: Found it: switching the trigram index to a covering index on repo_id and path. Starting with the migration.
One tool call, and it is caught up.
Sound familiar?
Re-explaining architecture decisions in every new chat
Cody starting fresh on a topic you covered last week
Decisions from your last session, gone
With MemoryRouter
search_memories and store_memory show up as MCP tools in Cody's agentic context
Ask it to check memory and it recalls the relevant history
Tell it to remember something and it is there next time
14-day free trial. Cancel anytime.
Connect Cody in five minutes
Three steps: enable MCP on your Sourcegraph instance, add the bridge, authenticate.
Turn on the agentic-context-mcp-enabled feature flag on your Sourcegraph Enterprise instance. If you are not an admin, ask whoever manages your Sourcegraph deployment.
{
"cody.mcpServers": {
"memoryrouter": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.memoryrouter.ai/mcp"]
}
}
}JetBrains uses the same cody.mcpServers block in cody_settings.json. You can also add a server from the Cody extension's MCP Settings UI (profile icon > MCP Settings).
The first time Cody calls a MemoryRouter tool, mcp-remote opens your browser to sign in and choose the vault it should use. Save a unique fact in one chat, start a new one, and ask Cody to check its memory for it. That is the moment it clicks.
Sign up free, then follow the three steps above.
How it works
Without MemoryRouter:
You Cody Model Response (gone when the chat ends)
With MemoryRouter:
You Cody (agentic context fetching) mcp-remote bridge MemoryRouter Model Response
mcp-remote runs locally as a plain stdio MCP server from Cody's point of view. It forwards every call over Streamable HTTP to MemoryRouter and handles the OAuth token so Cody never needs to. Your Cody account, model choice, and Sourcegraph search stay exactly where they are. MemoryRouter only stores and retrieves memories.
No key in settings.json
The bridge command carries no secret. OAuth handles sign-in the first time mcp-remote runs.
Explicit tools
search_memories, store_memory, and date_search_memories, each individually toggleable in Cody's MCP settings.
Same pattern other tools use
mcp-remote is the standard bridge for any stdio-only MCP client connecting to a remote, OAuth-protected server.
Same memory everywhere else
One vault. Connect it to Claude, ChatGPT, Cursor, or Claude Code and the memory follows you between tools.
Bridge it in five minutes. 14-day free trial.
Start free. Then just $20/mo.
14-day free trial. Cancel anytime.
- 200M memory tokens / mo included
- Unlimited projects
- Same memory in ChatGPT, Claude, Claude Code, and Codex
- 300 to 500 ms average recall on a typical vault
- Your keys, your data, your control
No inference markup, ever.
FAQ
Give Cody a memory tonight
14 days free. Five minutes to bridge. Tomorrow's first chat already knows everything from today.