AI Assistant Access (MCP)
Connect Claude or another AI assistant to your DeepSkyLog data over the Model Context Protocol
DeepSkyLog exposes your imaging data over the Model Context Protocol (MCP), so an AI assistant can answer questions about your projects, frames and equipment using your real data instead of guessing.
Access is read-only. An assistant can look at your data; it cannot change or delete anything.
What you can ask
Once connected, you can ask things like:
- “How many hours have I got on the Heart Nebula, and which filters?”
- “Which subs from my last M31 session had HFR above 3.5?”
- “What focal length and camera did I shoot IC 1396 with?”
- “What’s the cloud cover at my dark sky site over the next two nights?”
- “What should I image tonight?”
Connecting Claude
Open Claude, go to Settings -> Connectors, and choose Add custom connector.
Give it a name, for example
DeepSkyLog.Paste your MCP endpoint:
https://app.deepskylog.space/mcpClick Continue. Claude registers itself automatically - there is nothing to copy or configure.
You are sent to DeepSkyLog to sign in, then shown a consent screen naming the client and listing what it may read.
Choose Allow. You are returned to Claude, connected.
You can find the same endpoint inside the app under your name in the top right, then MCP access.
Claude Code and Claude Desktop
The same endpoint works from the command line:
claude mcp add --transport http deepskylog https://app.deepskylog.space/mcp
Your browser opens for the same sign-in and approval step.
Other assistants
The endpoint is a standard MCP server, so any client that supports remote MCP servers with OAuth should work. Clients differ in how much they support:
| Client | Status |
|---|---|
| Claude (web, desktop, Claude Code) | Supported and tested |
| ChatGPT developer-mode connectors | Expected to work, not yet verified |
| OpenAI Responses API | Expected to work, not yet verified |
| ChatGPT deep research connectors | Not supported - that mode requires differently named tools |
| Custom GPT Actions | Not supported - Actions use OpenAPI rather than MCP |
What the assistant can see
| Capability | What it covers |
|---|---|
| Projects | Names, targets, coordinates, status, integration time per filter, frame counts |
| Frame quality | HFR, FWHM, guiding RMS, eccentricity, star counts, quality scores and outliers |
| Equipment | Telescope, camera, mount, filter wheel, focal length, aperture, pixel size, filters |
| Locations | Coordinates, altitude and timezone of your observing sites |
| Weather | Forecast for your locations in three-hour steps |
| Tonight | Scored target recommendations from real visibility, moon separation and cloud cover |
Only your own data is reachable. Every request is scoped to the account that approved the connection.
Revoking access
Approval lasts until you withdraw it. Remove the connector in your AI client to stop using it. Access tokens are short-lived and expire on their own; if you want to be certain, contact support to have the connection revoked server-side.
Troubleshooting
The connector fails while checking the server. Confirm the URL ends in /mcp and starts with https://.
You are asked to sign in again. Access tokens are short-lived by design and the client renews them quietly. Being asked to approve a second time usually means the connector was removed and re-added.
The assistant does not see a new capability. Clients read the tool list once when they connect. Remove the connector and add it again to pick up newly added features.