cline

Engram connects to Cline through local MCP stdio. complete installation and the quick start, then use the absolute Python executable and config path printed by engram init. the connection does not depend on choosing Claude as Cline's model.

open the right settings

in the Cline editor panel, open MCP Servers → Configure → Configure MCP Servers. this opens the extension's own MCP settings JSON. use that file, rather than VS Code's separate .vscode/mcp.json; the editor profile and extension installation determine its actual storage location.

for Cline CLI, the documented file is ~/.cline/mcp.json. the cline mcp wizard can add or edit entries there. extension and CLI setup are described in Cline's official MCP guide.

add the local server

merge this into the file's existing mcpServers object:

{
  "mcpServers": {
    "engram": {
      "command": "/absolute/path/to/venv/bin/python",
      "args": [
        "-m", "engram",
        "--config", "/absolute/path/to/config.yaml",
        "serve", "--mcp"
      ],
      "disabled": false,
      "autoApprove": []
    }
  }
}

replace both paths. command names the executable; the array contains its arguments. the empty autoApprove list leaves individual tool approvals in place. this entry follows Cline's local-server schema.

Cline starts the server itself. do not start another copy in a terminal just to make the connection work, and do not enter the web dashboard's URL as an MCP endpoint.

environment and shared stores

the Python path must belong to the environment where Engram is installed. an editor process may not inherit your shell's virtual-environment activation. Windows JSON paths need escaped backslashes, such as C:\\Users\\you\\engram-env\\Scripts\\python.exe.

if you use hosted embeddings or Postgres, make the required environment variables available to the Cline process or configure the server's env mapping. keep secrets out of checked-in settings. use the same explicit Engram config when testing from a terminal.

all clients pointed at that config use its configured store. if two projects need separate memories, give their connections separate Engram configs and databases; changing the chat model does not switch the store.

check it from Cline

save the configuration, then enable or restart engram in MCP settings. confirm its tools appear and ask:

call Engram's status, then config_show, and show the connected store.

look for the actual tool response. Cline CLI can also show configured servers:

cline config mcp --json

this lists configuration; follow it with a tool call to verify the running connection. the official guide describes server management and CLI inspection.

troubleshoot

for connection errors, check Cline's MCP server error details and the absolute executable/config paths. for timeouts, run the local diagnostic first:

/absolute/path/to/venv/bin/python -m engram --config /absolute/path/to/config.yaml doctor --full

see doctor's checks and model behavior. restart the Cline server after fixing an error or upgrading Engram. when tools connect but recall is empty, recall_explain shows candidate scores and rejection reasons.

this guide uses stdio. Engram's optional remote transport is legacy SSE, not a generic Streamable HTTP endpoint.

source checked: official Cline MCP documentation, 20 september 2026. no editor-global storage path or client session is assumed.