Can Claude / ChatGPT / Cursor use Grants directly?

Yes, via the MCP server. It's a stdio JSON-RPC server that wires into Claude Desktop, Cursor, or any MCP-aware client in five lines of config. The list of tools matches your data model 1:1 - whatever a CRUD type exposes in the UI is exposed identically to the agent.

Do agent calls share the same auth + access controls?

Yes. The MCP server authenticates with the same JWT or PAT a human session uses. Sharing rules, org membership, rate limits, suspicion bans - all of it applies. There's no "agent backdoor" that bypasses the access matrix.

Is there an audit trail for agent actions?

Yes. Every mutation goes through the same audit-log path as a UI click. The transport (UI / API / CLI / MCP) is recorded as channel metadata so you can filter "what did the agent do this hour" from your account history.

Can I rate-limit an agent separately from a human?

Per-token rate limits apply, so you can issue an agent its own PAT and cap it independently. The shared user-level bucket also caps total throughput, so a runaway loop can't drain your whole account. The CLI's auto-update channel marker means you can also see in audit which version is calling.

Does the MCP / API surface cost extra?

No. Same plan, same surface. We don't paywall the agent - it's the same product, the same schema, the same access gates. If a plan can do X in the UI, it can do X via MCP.

Grants

Assess Directory Programmes Fairs

Compare

Grants: the agent-friendly alternative to ChatGPT / generic LLMs

Same product, same surface for humans and AI agents. MCP server + CLI + drop-in clients in 15 languages.

Agentic alternative

EXIST

ChatGPT

Finn GlasCo-Founder + Engineering

·January 22, 2026·

1 min read

Grants ships an MCP server you can wire into Claude Desktop or any MCP client in five lines. The same auth + access gates apply, the same audit log records every action, the same rate limits apply. ChatGPT / generic LLMs doesn't currently expose this surface - that's the gap.

At a glance

ChatGPT is unmatched at getting something on the page when you are blank. It is also unmistakably ChatGPT - reviewers spot the cadence in two paragraphs, the safety-rounded conclusions, the "moreover, furthermore" connective tissue. Grants carries the missing context: PTJ language conventions, EXIST review criteria, common rejection reasons (no academic anchor, innovation threshold not met, no LoIs from customers). Use ChatGPT to start; use us to finish.

Agent surface

What 'agent-friendly' actually means in practice

Grants ships an MCP server, a CLI, and drop-in clients in 15 languages. They're auto-generated from the same schema your UI uses; the auth, the access gates, the rate limits, the audit log all apply identically whether you're clicking buttons or calling the API. ChatGPT / generic LLMs doesn't currently expose this surface, which is the gap a serious agent integration trips over - the agent ends up scraping the UI through a browser, which is fragile, slow, and silently bypasses everything below.

Grants vs ChatGPT / generic LLMs: agent-readiness comparison

	Ours Grants	Theirs ChatGPT / generic LLMs
Knows what PTJ reads for
Knows EXIST review criteria
Generic-tone detection feedback
Free for unlimited drafts
Hosted in Germany / no US training
Strong at translating EN ↔ DE	Adequate

When to pick which

Pick Grants when

You're past the blank-page stage and ready for criteria-aligned editing.

EU/German data residency is required.

Pick ChatGPT / generic LLMs when

You have not started yet and need bulk text to react against.

You only need one or two passages translated, not the whole application reviewed.

Agentic-integration questions

What an agent-builder asks before wiring a tool into Claude / ChatGPT / Cursor.

Start with Grants

Free plan, no credit card. We host in Germany. Export + delete are self-serve.

Written by

Finn Glas

Co-Founder + Engineering

Finn is one of the Co-Founders. He owns the engineering side, the infrastructure, and most of the late-night fixes that ship before anyone notices.

finn.glas at aicuflow dot comLinkedIn Website