Tethys MCP/RAG

This is an MCP server that exposes LRAUV operator-assistant capabilities to any compatible client.

Status: Proof-of-concept

Static knowledge based on some WIP documents
A subset of LRAUV live information
Only query capability (no commanding or database changes)

Note

The server may be shut down or restarted at any time without prior notice.

The server combines static knowledge with live operational state (using TethysDash API), so a single client conversation can answer mixed questions like "Where is brizo right now and what does the manual say about emergency recovery?"

The MCP endpoint is available at (only accessible within the MBARI network):

http://tethystest.shore.mbari.org:8080/mcp

Connecting¶

Claude Code¶

claude mcp add tethys-ai --transport http http://tethystest.shore.mbari.org:8080/mcp

Start a fresh claude session and run /mcp, which should display tethys-ai as connected.

Claude Desktop¶

Through the GUI an https:// URL would be required. So, use a stdio bridge instead. You can edit your Claude Desktop config file and add the entry below.

Config file location:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "tethys-ai": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://tethystest.shore.mbari.org:8080/mcp",
        "--transport", "http-only",
        "--allow-http"
      ]
    }
  }
}

Restart Claude Desktop. The tethys-ai tools should appear in the tool picker.

Note

mcp-remote is a helper that runs locally and forwards the client (e.g., Claude Desktop) to the remote server. npx will fetch it on first run. --allow-http opts in to a plain-HTTP target (mcp-remote refuses non-HTTPS URLs by default).

Knowledge base¶

In the initial prototype only a couple of WIP documents have been ingested for Retrieval Augmented Generation (RAG).

Sample prompts¶

These are some basic prompts just for POC purposes.

Note

Most of these prompts were generated by Claude Code upon a simple request to use the implemented MCP tools as reference. So, some may be useless or unrealistic, but they should give a good general idea of the sort of questions that can be asked.

Knowledge base (RAG over ingested manuals/SOPs):

What does the operator manual say about watchstander duties?
What's the procedure for emergency recovery?
Show me the pre-deployment checklist.

Where & what (positions, missions, deployments):

Show brizo's positions over the last 12 hours.
What mission is makai running?
What's everyone running right now?
What's brizo's current deployment?
When was opah last deployed and what mission is it currently running?
Has makai started a new deployment in the last week?

Deployment documents (pre-deployment checklists, deployment plans, closeout reports):

What documents are attached to opah's current deployment?
Show me the predeployment checklist for pontus.
Who last filled out the closeout for opah's last deployment, and when?
Summarize the deployment plan for pontus.

Reachability (cell-comms ping):

Is brizo reachable by cell comms right now?
Which vehicles are currently reachable?

Events (operational log stream):

Any critical events on the fleet in the last hour?
What happened on brizo today?
Show me recent notes from opah.

Configuration:

What's been customized on opah?
What's opah's battery low-voltage threshold?
Compare brizo and opah's navigation tunings.

Time-series data (for plotting / trend reasoning):

What variables is daphne currently producing?
What science variables does ahi have right now?
Plot opah's depth over the last 4 hours.
Compare battery_charge between brizo and opah right now.
Has opah been at the surface recently?

Mixed-mode

Where is brizo right now and what does the manual say about emergency recovery?
Any critical events on opah today, and what does the runbook say about handling them?
Plot makai's depth and explain what its current mission is supposed to do.
Is opah currently doing what its deployment plan says it should?
Pontus's predeployment doc mentions CTD checks — is the CTD config customized on the vehicle?

Some exchanges via Claude¶

(See also here, here.)

NOTE: The following in no particular order, but captured in different quick sessions.

Brizo's current deployment includes taking ESP samples. Can you summarize its current status?¶

Brizo's current deployment ("Brizo 41 NOWRDC") launched August 3 and is actively underway — this is the one confirmed to be running ESP (Environmental Sample Processor) sampling.

Location/track: Working a patrol box off Monterey Bay, last GPS fix ~36.77°N, 121.86°W (as of 19:17 UTC today). It's been running a repeating "CircleSample" behavior — circling between 3–45 m depth to collect samples.

ESP sampling progress: 17 ESP samples completed so far this deployment. The two most recent:

Sample #17 (cartridge 40): filtered 3000 mL, completed normally ~18:43 UTC today
Sample #16 (cartridge 41): filtered 3000 mL, completed normally ~16:32 UTC today

Both logged "ESP sampling sequence completed normally," so the instrument is cycling through cartridges and filtering water as planned. Each sample logged a recurring but apparently benign message ("AnalyticModule::Error in Probe-AMspr — No Response to BREAK") before cartridge selection — it hasn't stopped any sample from completing, so it looks like routine chatter rather than a fault.

Health: No faults, critical logs, or emergencies in the recent event stream. Ground-fault scan came back clean. Comms are healthy — ping checking is enabled and the vehicle is sending frequent Iridium/SBD packets and GPS fixes (every ~15–30 min).

Bottom line: Brizo is deployed, healthy, and actively executing its ESP sampling mission with no red flags.

What's the general status of the LRAUV fleet at the moment?¶

Fleet snapshot as of today (July 28):

Currently at sea: Triton is the only vehicle actually in the water — deployed July 27, still out on mission "Triton 59 TritonCam."

Everything else is on the bench, with recent history:

Makai — recovered July 23 (5 days ago), completed a 10-day "Piscivore" mission
Ahi — recovered July 22 (6 days ago), "mbts" mission
Pontus — recovered July 9 (3 weeks ago), i2map mission
Daphne — recovered June 25 (~1 month ago), Sipper test
Opah — recovered June 9 (~7 weeks ago)
Galene — just finished a tank test (not an ocean deployment), ended yesterday
Tethys — last actual deployment was January 2024, so it's been out of ocean rotation for a while (likely used for dock/bench testing rather than missions currently)
Aku — last out in December 2025
Melia and Pallas — no recent activity at all; their last logged deployments are from 2013 and 2019, so these look retired or inactive
Brizo — has a placeholder deployment queued for early August, but hasn't launched yet (upcoming, not current)
Sim — a simulation/testbed entry that's continuously "deployed" for testing, not a physical vehicle
Mesobot, Polaris, Proxima, Pyxis, Stella — no deployment records found in the system at all

So of the 18 named vehicles tracked, exactly one (Triton) is genuinely underway right now, with a cluster of others (Ahi, Makai, Pontus, Daphne, Opah) having cycled through the water in just the past two months — suggesting a fairly active recent cadence even though only one is out today.

Is Triton currently deployed and if so, what's its basic status?¶

Yes, Triton is currently deployed. Quick summary:

Triton is on deployment "Triton 59 TritonCam - Deployable AI" (git tag 2026-07-27_A),
launched July 27 at 11:22 AM PDT and still in the water — no recovery or end time logged yet.
Most recent GPS fix was at 2026-07-28 19:30 UTC (~12:30 PM PDT today),
positioned around 36.798°N, 121.869°W, just off Monterey Bay,
following a survey pattern (recent waypoints include "expanding_donut" and "adaptive_yoyo"
mission components). Cell-comms ping checking is enabled and mostly reachable,
with occasional brief dropouts (one failed ping in the last few checks, otherwise reachable).

What mission is makai running?¶

⏺ Makai is currently running the Default mission (started 2026-04-29 22:00:15 UTC).

What's rizo's current deployment?¶

⏺ Brizo's most recent deployment is Brizo 36 ESP/MARL (2025-12-17_B):

Launched: 2025-12-08 16:53 PST
Recovered: 2025-12-18 09:46 PST
Note this deployment ended in December — brizo is not currently in the water.

Has opah started a new deployment in the last week?¶

⏺ No — opah's most recent deployment is Opah 17 with git tag 2026-04-10, which started around April 10, 2026 (about 19 days ago), so well outside the last week.

What vehicles are being monitored for cell connectivity?¶

⏺ Two vehicles have cell-comms ping checking enabled:

brizo
pontus
All other vehicles in the fleet (ahi, aku, daphne, galene, makai, melia, mesobot, opah, pallas, polaris, proxima, pyxis, sim, stella, tethys, triton) have ping checking disabled.

What's opah's battery low-voltage threshold?¶

⏺ Opah's battery low-voltage threshold is IBIT.batteryVoltageThreshold = 13.7 V — "Amount of voltage left at which we return to the surface" (default, not customized).

Related thresholds for context:

BPC2.batteryWarningVoltageThreshold = 14.5 V (warning level)
IBIT.batteryCapacityThreshold = 50 Ah (return-to-surface charge level)

Plot opah's depth over the last 4 hours.¶

⏺ I can't render a real plot in the terminal, but here's the data summary — opah is essentially flat at the surface (no diving):

Depth (m) over 2026-04-29 19:00 → 23:00 UTC (28 unique samples)

Min: 0.060 m
Max: 0.092 m
Range: ~3 cm of variation
Trend: very slight downward drift (0.092 → 0.064 m over the 4 h)
ASCII sketch (each █ = a sample, vertical scale 0.06 → 0.10 m):

  0.10 ┤
  0.09 ┤██████████
  0.08 ┤          ██████████
  0.07 ┤                    ██████
  0.06 ┤                          ████
       └────────────────────────────────
       19:00      20:00      21:00      22:00 UTC
Opah isn't diving — depth holds within a 3 cm band near zero. This is consistent with the vehicle being on deck/at the surface rather than on a deployed mission profile. Want me to pull a different variable (e.g. battery_charge, sea_water_temperature) or widen the time window?

Where's pikachu at the moment?¶

⏺ There's no vehicle named pikachu in the TethysDash fleet.
The known vehicles are: ahi, aku, brizo, daphne, galene, makai, melia, mesobot, opah, pallas,
polaris, pontus, proxima, pyxis, sim, stella, tethys, triton.

Did you mean one of those? (e.g. pyxis or pallas?)

Summarize the latest pre-deployment document for Ahi¶

⏺ Latest pre-deployment doc for Ahi
"Ahi 50 - Synchro OSW - Predeployment Checklist", filled out by [MASKED] on 2026-05-18 (≈4 h ago).
Tied to active deployment Ahi 50 Synchro OSW (id 3000742).

Header

Date of last pressure-vessel close-out: 2026-01-21
Marked as attached to deployment ✓
Standard 22-item checklist — all major items ticked
Including: deployment started on Dash with matching tag, logs synced, system clock set UTC, tools disabled,
SBIT/IBIT pass, burnwire ~24 V verified, tether disconnected, drop weight
installed, aft fairing on, antenna deflector confirmed, AT5 acoustic tracking beacon, wildlife tag activated,
payload functional, Iridium + ARGOS (status 127) transmissions received.

IBIT sub-checks — all confirmed: stop/abort depths, battery thresholds, neutral buoyancy value,
battery capacity adequate, persisted values, compass orientation matches.

Operator note (from the IBIT section): "Nano beacon 2311 installed and charged to 95%. tested with iWand 14MAY2026 - cwahl"

Items not ticked

The four pre-checklist "starting state" boxes (aft fairing off, control surfaces on, propeller on,
Sonardyne beacon installed) — were they assumed-true rather than explicitly confirmed?
Worth a glance.
"Verify ESP scripts run (if ESP installed)" — and both ESP-notes checkboxes at the bottom are blank,
consistent with this deployment not carrying an ESP.
Bottom line: the vehicle was signed off as ready to launch. The only gaps are the four unticked preconditions;
everything ESP-related was skipped, which is consistent with no ESP onboard.