Using the Observatory with AI Assistants

Connecting your client

Add the Observatory endpoint to your MCP-compatible client. Configuration format varies by client — examples for common clients are below.

[mcp_servers.faultline]
url = "https://mcp.faultlinewatch.com/mcp"

Settings → Connectors → Add custom connector

https://mcp.faultlinewatch.com/mcp

{
  "mcpServers": {
    "faultline": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.faultlinewatch.com/mcp"]
    }
  }
}

# Transport Type: Streamable HTTP
# URL:
https://mcp.faultlinewatch.com/mcp

claude mcp add --transport http faultline https://mcp.faultlinewatch.com/mcp

curl -s https://mcp.faultlinewatch.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "faultline_search_records",
      "arguments": { "query": "room-temperature superconductivity", "limit": 1 }
    }
  }'

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "{\n  \"query\": \"room-temperature superconductivity\",\n  \"count\": 1,\n  \"records\": [\n    {\n      \"id\": \"FR-AM-0005\",\n      \"programme\": \"PROG-AM\",\n      \"programmeName\": \"Advanced Materials, Physics & Energy\",\n      \"claim\": \"Room-Temperature Superconductivity — Reproducibility Under Laboratory Conditions\",\n      \"status\": \"closed\",\n      \"pressureState\": \"collapsed\",\n      \"verificationStage\": \"VS-04\",\n      \"assessmentDate\": \"2026-06-29\",\n      \"openedDate\": \"2024-01-15\",\n      \"lastMutationDate\": \"2026-07-09\",\n      \"evidenceInstances\": 6,\n      \"assessments\": 2,\n      \"openQuestions\": 4,\n      \"canonicalUrl\": \"https://faultlinewatch.com/the-record/fr-am-0005/\"\n    }\n  ]\n}"
      }
    ]
  }
}

Access is public and read-only. No API key, account, or authentication step is currently required. Most users can connect directly using the endpoint shown above. Reasonable rate limits may apply to protect service availability, particularly for large cross-record queries or automated loops.

The /mcp endpoint is a single stateless POST route — every call is a self-contained JSON-RPC request (no session handshake or cookie state is kept between calls), which is what makes the raw HTTP example above work without an MCP client library.

faultline_list_records and faultline_search_records both accept an optional detail parameter: "summary" (default, unchanged) returns one thin projection per record — the same shape always returned before this parameter existed — while "full" returns the same complete canonical view as faultline_read_record for every matched record, so a cross-record query can retrieve evidence instances, mechanisms, and open questions in a single call instead of onefaultline_read_record call per hit.

Where structured provenance has been recorded, an evidence instance can expose a canonical sources[] array. Each source has a citation and may also provide a URL, DOI, locator, or short exact quote. One instance may cite several sources. Legacy instances may legitimately have no sources[]; that means structured provenance has not yet been recorded for that instance, not that the evidence has no source.

Once connected, your client discovers the Observatory tools automatically. The machine interface derives from the canonical corpus rather than maintaining a separate case list, so a governed record change has one institutional source regardless of whether it is read on the website or through MCP.

After adding the endpoint, restart your client, confirm that the Faultline Observatory tools appear in the available tool list, and begin a new chat session. Start with a simple record lookup before moving to corpus-level analysis.

Try

• Read FR-AM-0005 and explain why its current Pressure State remains Collapsed.
• What evidence instances are recorded for FR-AI-0009?
• What sources support the evidence instances in FR-AI-0001, and which can I inspect directly?
• Which PROG-AI records are currently Fragmenting?
• Search the corpus for records that mention reopening conditions.

Cross-record analysis

Most visitors use the website to read individual Frontier Records. The MCP connection enables an AI assistant to retrieve and compare canonical records directly, while preserving the distinction between evidence and institutional assessment.

For example:

• Which records have accumulated the strongest contradictory evidence?
• Which records have remained stable for long periods without reassessment?
• Which programmes contain the largest number of unresolved open questions?
• Where has new evidence been added without changing the institutional judgment?

The assistant can inspect the underlying evidence instances, their recorded source provenance where available, assessment history, and mutation record used to support such comparisons.

What the records do not do