# Attify for developers

> For developers connecting an AI agent to Attify: the Ingest API, Python and TypeScript SDKs, OpenTelemetry, platform connectors, and the read-only API and MCP server agents use.

Published 2026-11-12 · Updated 2026-11-17 · Reviewed 2026-11-12. Canonical: https://attify.app/for/developers

You can send an agent's interactions to Attify in four ways (connect the customer's systems, connect an agent platform or observability account, stream through the Ingest API or an SDK, or export OpenTelemetry traces), and read audited results back through a read-only API, llms.txt and an MCP server. The integration carries three things Attify cannot infer: the agent's claimed outcome, any handoff, and the customer's own ids.

## Send

| Path | What you do |
|---|---|
| Customer's systems | Nothing: the customer connects Intercom, Zendesk, HubSpot, Salesforce, Gorgias, Freshdesk, Google Calendar, Microsoft 365, Calendly or Cal.com |
| Agent platform or observability | Paste a key (read-only where the platform offers one) for Vapi, Retell AI, ElevenLabs, Bland AI, Voiceflow, Langfuse or LangSmith |
| SDKs | `pip install attify-ingest` and `npm install attify-ingest`; `interaction()`, `message()`, `claim()`, `end()` |
| Ingest API | `POST /v1/ingest/events` (JSON batches of 1 to 500) or `/v1/ingest/stream` (NDJSON), signed |
| OpenTelemetry | Point the OTLP exporter at `/v1/otlp/v1/traces`; set `proven.outcome` on the span |

Event schema `proven.ingest/1`, JSON Schema at `/docs/ingest/schema.json`, OpenAPI at `/docs/ingest/openapi.json`. Full reference: [/docs/ingest](https://attify.app/docs/ingest). Walkthrough: [how to send your agent's data](https://attify.app/guides/send-agent-data-to-attify).

## What to carry

- `outcome.claimed` in the job type's words: support `resolved` / `not_resolved`; sales meetings `meeting_booked` / `no_meeting`; bookings `booked` / `not_booked`. The latest claim is the one tested.
- `handoff` with `to`: `human`, `agent` or `external`.
- `customer` and `external_ref`: ids in the customer's own systems, so repeat contacts and spot checks can be found there.
- `feedback` of kind `resolution` only for an explicit "did this resolve your issue?" answer; ratings go as `csat`, `thumbs` or `nps`.

## What Attify checks

Signatures, timestamps on arrival (events more than 5 minutes late are flagged, and their interactions never scored as witnessed), sequence gaps and reused ids. Witnessed data counts toward verification only after the customer consents and [reconciles](https://attify.app/glossary/reconciliation).

## Read

- `GET /api/v1/vendors?job_type=support-resolution`, `GET /api/v1/vendors/{slug}`, `GET /api/v1/index`.
- `POST /api/v1/verify` with an attestation; keys at `/.well-known/proven-keys.json` (JWKS, Ed25519).
- MCP (Streamable HTTP, `POST /mcp`): `lookup_vendor`, `verify_attestation`, `get_traction_report`, `get_index`, `explain_provenance`.
- `/llms.txt` for agents that start there.

## Questions

### Is there an SDK?

Yes: attify-ingest for Python (pip install attify-ingest, standard library only) and TypeScript (npm install attify-ingest, no dependencies). Both sign, number, buffer and retry; nothing else. Any HTTP client works too.

### How are requests authenticated?

With the deployment's key: HMAC-SHA256 of the timestamp and the raw body in the Proven-Signature header for /v1/ingest/events, or a bearer token for OTLP. Timestamps more than 300 seconds off are refused.

### Can another agent read audited results?

Yes: the read-only API at /api/v1 (OpenAPI at /api/v1/openapi.json), llms.txt, and an MCP server at /mcp with tools such as lookup_vendor, verify_attestation and get_index.

## Sources

- Attify: [Ingest API, schema proven.ingest/1](https://attify.app/docs/ingest), read 2026-11-08
