A quick MCP server to give AI access to a database
Capability for LLMs to write an observation log, exposed as an MCP server. SQLite backend, Bearer token auth, Streamable HTTP transport.
location /mcp {
proxy_pass http://127.0.0.1:8742;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
}Get a cert on it (certbot/acme). The MCP endpoint is https://yourdomain.com/mcp.
In Claude → Customize → Connectors → Add custom connector:
https://yourdomain.com/mcpAPI_SECRET valueClaude passes it as Authorization: Bearer <secret> on every request.
| Tool | Description |
|---|---|
add_observation | Write a new observation. Requires summary (≤50 chars), observation, note. |
list_observations | List all entries newest-first. Returns (id, summary, created_at) only. |
search_observations | LIKE search on summary + observation text. Returns (id, summary, created_at). |
get_observation | Fetch full entry by ID. |
update_observation | Patch any fields on an existing entry by ID. |
delete_observation | Delete entry by ID. |
Search returns lightweight rows intentionally — call get_observation(id) once you have the right ID.
id INTEGER autoincrement
summary TEXT ≤50 chars, used for retrieval
observation TEXT full observation
note TEXT sentiment / interpretation
created_at TEXT UTC ISO 8601| Variable | Required | Description |
|---|---|---|
API_SECRET | ✅ | Bearer token. Server refuses to start without it. |
DB_PATH | ❌ | Path to SQLite file. Defaults to /data/mira.db. |
ieepirzy/databse-MCP
April 11, 2026
April 13, 2026
Python