Connect your Claude / GPT
Your Claude or ChatGPT, connected to An:des
AndesLex is the An:des app: the name of the connector and of your account.
These steps are for Claude. In Your systems and AI, you can explore the available connections for other assistants.
What you need
- A Claude account on a plan that lets you add custom connectors
- Your AndesLex account
Add the connector
- In Claude, open Settings → Connectors → Add custom connector
- Name: AndesLex. URL: https://andeslex.cl/api/mcp
- Press Connect, sign in with your AndesLex account and review the consent: it states which programme you can read and until what date
Try it
- «Which programmes do I have in AndesLex?»
- «In the test programme, what is missing and who must resolve it?»
- «What capabilities does the system have?»
- «Review the screening of the sample supplier.»
What it does and what it does not
- Claude reads only what your delegation authorises, and every read is logged in An:des
- It does not approve or sign. Approvals stay with whoever holds the authority within your organisation
- The delegation has an end date and can be revoked
How An:des governs your AI
- Investiture: your AI acts only with a delegation in force, granted by whoever holds the authority, with an end date and revocable.
- Competence: only the programmes and services the delegation names; everything else is out of scope.
- Form: every read and every contribution is logged, with its PoD, and Argos verifies before anything goes out.
This is how An:des works with Érika and with any AI you connect: if the delegation is not current or the action falls outside what it names, the app rejects it; what it does is recorded. Users, credentials, access and credits, each in its own place.
You decide how much it does on its own
- Explains: your position, your risks and what is missing
- Suggests: the next steps, with their reason
- Prepares: drafts that do not go out without your approval
- Executes: bounded, low-risk tasks, logged and reversible; never acts
Anything high-risk always goes through whoever holds the authority, and an emergency stop halts it all. An:des communicates only with whoever you authorise, under the rules you approved beforehand. That way, what the AI reads and reports in your case is recorded, with the delegation that authorised it.
Twelve services, in three governances and one transversal
Existence: read the programme’s status and its list screening, with matches and sources. Absence: read the work, with its pending items, and submit a report. Artifact: read a PoD receipt, prepare and track a document batch, and continue its intake, verify each file and request the close. Transversal: list your authorised programmes and the system’s capabilities. Every service requires the delegation in force; a read grants no authority and a report does not decide.
Measured in the app on 25 Sept 2026: Sentinel detects in 4 seconds that a rule’s validity no longer matches its source. Érika retrieves an answer with its citations in 52 seconds. You are in the real agent ecosystem: the same connector serves Claude, ChatGPT and your own code.