GetItDone for AI agents
How AI assistants and agents read and build against GetItDone: llms.txt, Markdown twins of every docs page, the MCP server, and agent guidelines.
Everything an AI assistant or agent needs to read our docs and build against the API — in machine-friendly formats.
Machine-readable indexes
llms.txt— a compact index of the site and developer surface.llms-full.txt— the entire developer docs set as one Markdown document.openapi.json— the OpenAPI 3.1 document for the whole/v1surface.
Markdown twins
Append .md to any developer docs URL to get its Markdown source — ideal for
feeding a page straight into a model.
https://nowgetitdone.com/docs/api.md
https://nowgetitdone.com/docs/ai.md
https://nowgetitdone.com/docs/api/webhooks.md
https://nowgetitdone.com/docs/api/problems/rate-limited.mdConnect over MCP
The GetItDone MCP server lets an assistant work with tasks directly — including image and video-frame context. Endpoint:
https://app.nowgetitdone.com/api/mcpOne platform: REST + SDK + MCP
The REST API, the TypeScript SDK, and the MCP server are three doors into the SAME platform — one credential, one set of scopes, one docs surface. The SDK is a typed client over the REST API, so its capabilities track REST exactly. The MCP server (10 tools) is a focused, agent-facing subset plus one capability REST deliberately does not expose.
| Capability | REST /v1 | SDK | MCP |
|---|---|---|---|
| Tasks — list, create, update, archive | Yes | Yes | Yes |
| Task full content + attachment URLs (read) | Yes | Yes | Yes |
| Workspaces / organizations (read) | Yes | Yes | Yes |
| Task history / versions | Yes | Yes | No |
| Daily plan | Yes | Yes | No |
| Projects | Yes | Yes | No |
| Members | Yes | Yes | No |
| Attachments — upload / download | Yes | Yes | No |
| Outbound webhooks | Yes | Yes | No |
| Usage & quota | Yes | Yes | No |
| Video-frame task context MCP-only by design (see divergences) | No | No | Yes |
| Docs search Agent convenience over the public docs pages | No | No | Yes |
Documented divergences
get_task_video_contextis intentionally MCP-only. Extracting frames from a task's videos so an agent can see them is an MCP-native rich-context capability; exposing raw extracted frames over REST is deliberately out of scope for v1. This is a documented divergence, not a missing feature.- The GitHub integration tools are copilot-only. They power the in-app AI
copilot and have no public REST surface; adding GitHub actions to
/v1is a deliberate, owner-gated scope expansion — not a gap v1 intends to fill.
REST quickstart
Authenticate with a gid_ key over Bearer auth:
curl https://app.nowgetitdone.com/v1/organizations \
-H "Authorization: Bearer gid_YOUR_API_KEY"TypeScript SDK
A typed client is available — npm install @nowgetitdone/sdk
(source on GitHub).
Agent behavior guidelines
- Respect the scopes on the credential — request only what you need, and
expect
403 insufficient_scopeotherwise. - Send an
Idempotency-Keyon consequential POSTs so retries are safe. - Back off on
429responses, honoring theRetry-Afterheader. - Switch on the stable
codein problem responses, never the human title.
Replay all recently failed deliveries POST
Bulk sibling of the single replay: creates FRESH PENDING deliveries for this endpoint's FAILED/DISABLED deliveries created at or after `since`, skipping every event that already has a PENDING/DELIVERING delivery queued for this endpoint (no double-enqueue) — hard-capped at 1000 replays per call (call again with the same `since` for the rest; already-replayed events are skipped by the pending check). Repeating the call is naturally near-idempotent for the same reason, so this route does not use the Idempotency-Key ledger. Does NOT re-enable a disabled endpoint — the “re-enable & replay” console flow is: update `enabled: true` (which resets the failure counter), then call this.
Pricing
GetItDone plans — Free, Pro, and Team.