Generate one hosted MCP secret, run Supermemory locally, then point your coding agent at the ErrorAtlas MCP server. No repo clone, database setup, or Worker deploy needed.
Open the hosted dashboard, sign in, click Generate secret, then copy it immediately. This secret authenticates every MCP request. Existing raw secrets cannot be recovered later.
https://erroratlas.sansynx.workers.dev/dashboard
Install ErrorAtlas in the project where your coding agent works.
npm install -D @sansynx/erroratlas npx erroratlas init
Supermemory stores private debugging context on the developer machine. It needs one model-provider key such as OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, or GROQ_API_KEY. Read the official Supermemory quickstart if you need provider setup help.
OPENAI_API_KEY=your_model_provider_key npx supermemory local --port 6767 # if installed globally supermemory local --port 6767
Keep this terminal running while your coding agent works. Local memories are stored in the project's .supermemory folder.
Windows note: if the npm wrapper says the server binary was not found, open bash or WSL in the project and run PORT=6767 ~/.supermemory/bin/supermemory-server after exporting your model-provider key.
Paste this into your agent client's MCP config. Replace ERRORATLAS_API_KEY with the secret copied from the dashboard.
{
"mcpServers": {
"erroratlas": {
"command": "npx",
"args": ["-y", "--package", "@sansynx/erroratlas", "erroratlas", "mcp"],
"env": {
"ERRORATLAS_API_URL": "https://erroratlas.sansynx.workers.dev",
"ERRORATLAS_API_KEY": "ea_live_from_dashboard",
"SUPERMEMORY_URL": "http://localhost:6767"
}
}
}
}
Ask the agent to search before fixing, capture meaningful failures, and publish a playbook only after the fix is verified.
- Before changing code, search ErrorAtlas. - If the error is useful, call capture_error_signal. - If the fix works, call publish_resolution with root cause, fix, and verification.