Connect OpenCode or Pi
Set MEMINI_READ_TOKEN in the shell where you will launch
the client. Use a read token scoped to the demo workspace.
The shared setup includes Bash and PowerShell commands. Prepare your token
OpenCode
Run opencode --version to choose the config layout. In
your project’s opencode.jsonc, merge the following entry
with any MCP servers you already use. OpenCode expands
{env:MEMINI_READ_TOKEN} when it loads the file.
OpenCode 1.x uses the server name directly under
mcp. OpenCode
1.x MCP configuration
{
"mcp": {
"memini": {
"type": "remote",
"url": "http://127.0.0.1:6364/mcp",
"headers": { "Authorization": "Bearer {env:MEMINI_READ_TOKEN}" },
"oauth": false
}
}
}OpenCode 2.x places the same server entry under
mcp.servers. OpenCode MCP
configuration
{
"mcp": {
"servers": {
"memini": {
"type": "remote",
"url": "http://127.0.0.1:6364/mcp",
"headers": { "Authorization": "Bearer {env:MEMINI_READ_TOKEN}" },
"oauth": false
}
}
}
}Run opencode mcp list from the project directory. Look
for memini marked connected, then launch
opencode from that shell.
Pi
Pi has a built-in Streamable HTTP client. Add the entry to your
personal ~/.pi/agent/mcp.json, or use
.pi/mcp.json in a trusted project. Pi expands
${MEMINI_READ_TOKEN} in the header. Pi MCP configuration
{
"mcpServers": {
"memini": {
"url": "http://127.0.0.1:6364/mcp",
"headers": { "Authorization": "Bearer ${MEMINI_READ_TOKEN}" }
}
}
}Run pi mcp list. A successful connection lists
memini: connected and tools including status
and search. Start pi in that shell. Use
/mcp inside Pi to inspect the server or reconnect it.
Check workspace access
Ask the agent: “Call the Memini status tool and show me
the index status for workspace demo.” If the connection
returns HTTP 401, reload the read token in the client’s shell and
restart the client. If the connection is refused, check that the Memini
server is running and that its address is reachable from the client. Memini connection check