One endpoint, every check

Authentication

Create a personal access token under Settings → API tokens and send it as a bearer token. Tokens carry the cli scope and are tied to your account's credits. Anonymous calls get a 401.

curl https://truestandard.ai/api/v1/me \
  -H "Authorization: Bearer ts_live_…"

Check types

This table is read from the server at render time, so it is the catalog as of now. GET /api/v1/checks returns the same list as JSON; any alias below works in the URL.

GET /api/v1/checks

Depths

You pick the depth on every call. Nothing is auto-detected.

Run a check

POST to /api/v1/checks/:type with input and depth. Council and deep need credits; single needs nothing but a token. This response is a real single run of the claim check, copied from the server unedited.

Request

curl -X POST https://truestandard.ai/api/v1/checks/claims \
  -H "Authorization: Bearer ts_live_…" \
  -H "Content-Type: application/json" \
  -d '{"input": "The Berlin Wall fell in November 1989, and East and West Germany formally reunified the following year.", "depth": "single"}'

Response (200)

{
  "status": "completed",
  "tool": "claim-checker",
  "depth": "single",
  "mode": "single_model",
  "data": {
    "claim": "The Berlin Wall fell in November 1989, and East and West Germany formally reunified the following year.",
    "verdict": "verified",
    "confidence": "high",
    "explanation": "The Berlin Wall fell on November 9, 1989. Formal reunification (German Unity Day) took place less than a year later on October 3, 1990. The timeline is spot on.",
    "publish_safe": true,
    "correct_version": null,
    "shareability_risk": "low",
    "source_assessment": "No source is explicitly cited in the claim, but the dates are historically documented and undisputed facts."
  },
  "models": [
    "Gemini"
  ],
  "duration": 1.77,
  "receipt": {
    "id": "JV0k-5ap",
    "url": "https://truestandard.ai/check/JV0k-5ap",
    "shared": false
  },
  "remaining": 0
}

Deep runs are polled

A deep call answers 202 with an id and a status URL. Poll it every few seconds until status is completed or failed; the completed payload is the same shape as a synchronous run, with credits_used and balance filled in.

POST /api/v1/checks/premortem   {"input": "…", "depth": "deep"}
→ 202 {"status": "pending", "id": "EaSCTdqM", "tool": "pre-mortem-generator", "depth": "deep",
       "status_url": "https://truestandard.ai/api/v1/checks/EaSCTdqM"}

GET /api/v1/checks/EaSCTdqM
→ 200 {"status": "processing", …}        keep polling
→ 200 {"status": "completed", "data": {…}, "models": [...], "credits_used": 3, "balance": 196, "receipt": {…}}
→ 200 {"status": "failed", "error": "…"}  nothing charged

Review: the composed check

POST /api/v1/checks/review sends a whole draft through the claim and citation checks, and returns one verdict with a publish-safe count and a public report. One check type per run: for slop, sycophancy or a premortem, call that check on its own. Council or deep only. It answers 202; poll GET /api/v1/verifications/:id. At deep the server may first ask clarifying questions; POST /api/v1/clarifications returns them.

POST /api/v1/checks/review
{"content": "…the whole draft…", "depth": "council", "context": "creator"}
→ 202 {"id": "…", "status": "pending", "depth": "council"}

GET /api/v1/verifications/:id
→ 200 {"status": "completed", "verdict": "…", "publish_safe": {"safe": 7, "total": 9},
       "consensus_score": 0.78, "model_responses": [...],
       "credits_used": 1, "balance": 198}

Errors

Every error is { error, message } plus context fields. The error string is what the CLI turns into an exit code, so an agent can branch on it without parsing prose.

From the CLI

The CLI is the same API with auth and polling handled for you. JSON goes to stdout, progress to stderr.

curl -fsSL https://truestandard.ai/install.sh | sh
truestandard auth login
truestandard check --list
truestandard check claims "The Berlin Wall fell in November 1989." --depth single
truestandard check citations --file draft.md --depth council
truestandard check review --file draft.md --depth deep

CLI

From Claude Code, Cursor and other MCP agents

truestandard mcp serves the checks as MCP tools over stdio: check_claims, check_citations, check_slop, check_sycophancy, premortem, red_team, compare_answers, review, a generic check and list_checks. Each returns the JSON above as text, receipt URL included. An out-of-credits or rate-limited call, or one without a token, comes back as a tool error that says what to do, so the agent stops instead of retrying.

claude mcp add truestandard \
  -- truestandard mcp

{
  "mcpServers": {
    "truestandard": {
      "command": "truestandard",
      "args": ["mcp"]
    }
  }
}

[mcp_servers.truestandard]
command = "truestandard"
args = ["mcp"]