Connect Claude to Silent Witness

Connect Claude to Silent Witness and you can ask about a case in plain language — "what was the delta-v?", "what do the medical records say?", "how strong is liability?", "is my report ready?" — and get an answer drawn from that case's real analysis results.

Claude reads your cases. It cannot change them: nothing you ask in a conversation can create a case, upload or delete a file, start an analysis, or generate a report. Those stay in the Silent Witness web app.

Setup takes about five minutes.

What you can ask

Once connected, Claude can see, for any case on your account:

Case details	parties, side, analysis type, accident date and location
Vehicles	year, make, model, weight, number of impacts, whether EDR data exists
Occupants	seating position, demographics, alleged injuries
Crash reconstruction	delta-v, PDOF, crash pulse and peak acceleration for each impact
Case intelligence	liability assessment, claim exposure, case strength
Biomechanics	per-occupant results and the generated opinions
Evidence	what has been uploaded — file names, types and processing status
Medical records	what the extraction found: diagnoses, treatment, providers
Report	whether the technical report is ready and where to download it

If an analysis has not been run or is still running, Claude will tell you that rather than guess. "The biomechanics analysis hasn't finished yet" is a real answer you should expect to see.

Generated findings and opinions are drafts prepared for attorney review — not filed expert opinions, and not legal advice.

Connect from claude.ai

1. Open claude.ai and go to Settings → Connectors.

2. Choose Add custom connector.

3. Enter the server URL:

   https://mcp.silentwitness.ai/mcp

4. Click Connect. A Silent Witness sign-in page opens in your browser.

5. Sign in with your normal Silent Witness account and approve the request to read your cases.

You are done. Start a conversation and ask about a case by name — "what do we know about the Martinez case?" — and Claude will find it and answer.

Connect from Claude Desktop

1. Open Settings → Connectors → Add custom connector.
2. Use the same URL: https://mcp.silentwitness.ai/mcp
3. Sign in when the browser window opens.

Connect from Claude Code

claude mcp add --transport http silentwitness https://mcp.silentwitness.ai/mcp

Run /mcp inside Claude Code to complete sign-in.

Connecting with an API key instead

If you are building an integration, or your environment cannot open a browser for sign-in, authenticate with a Silent Witness API key instead.

Create one in the web app under Settings → API Keys. The key needs the read scope, which is the default for new keys.

Send it as a bearer token:

curl https://mcp.silentwitness.ai/mcp \
  -H "Authorization: Bearer sk-your-key-here" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

An x-api-key: sk-… header works too.

In a client config file:

{
  "mcpServers": {
    "silentwitness": {
      "type": "http",
      "url": "https://mcp.silentwitness.ai/mcp",
      "headers": { "Authorization": "Bearer sk-your-key-here" }
    }
  }
}

Treat an API key like a password: it grants read access to every case on your account. Sign-in through the browser is the better option whenever it is available.

What Claude can see

Only your account's cases. A case belonging to another firm is not readable and not discoverable — asking for one by ID returns "case not found", the same answer you get for an ID that does not exist.

Every request is logged: which account, which tool, which case, and what the outcome was. The log records identifiers only, never the case content itself.

Medical records are protected health information. Claude answers the question you asked and does not volunteer unrelated medical detail — but the conversation itself lives in your Claude account, so treat it the way you would treat any other place you discuss case material.

Troubleshooting

"Insufficient scope" or a 403. Your API key does not carry the read scope. Check it under Settings → API Keys; a key created without an explicit scope already has read access.

"Case not found" for a case you know exists. The case belongs to a different account than the key or login you connected with. Check which account you are signed in as.

Claude says an analysis isn't available. That is usually accurate — the analysis has not been run, or is still running. Open the case in the web app to confirm and start it if needed.

Answers seem out of date. If evidence changed after the last calculation, Claude will say the results may be stale. Re-run the analysis in the web app.

Too many requests. Each credential has a request limit. An ordinary conversation will not reach it; a script might. Slow down and retry.

Limits

• Read-only. No case creation, uploads, deletions, analysis runs, or report generation through Claude.
• One case at a time. Claude answers about a specific case; it does not run comparisons across your whole caseload.
• Files stay in Silent Witness. Claude sees that a photo or document exists and its processing status, never the file itself.