API reference

Base URL: https://mcp.atlaso.ai/developer/v1. Authenticate with a project key + subject, or a device bearer — see Authentication. All bodies are JSON.

Remember

Store a memory in a subject’s bag. POST /memories Body

Field	Type	Notes
`text`	string	Required. 1 byte – 64 KB UTF-8.
`tags`	string[]	Optional. Up to 32 tags, each ≤128 chars.

Pass an optional Idempotency-Key header to make retries safe (details).

curl -X POST https://mcp.atlaso.ai/developer/v1/memories \
  -H "X-Atlaso-Project-Key: $ATLASO_KEY" \
  -H "X-Atlaso-Subject: user_8213" \
  -H "Content-Type: application/json" \
  -d '{"text": "Prefers window seats.", "tags": ["prefs"]}'

import requests

requests.post(
    "https://mcp.atlaso.ai/developer/v1/memories",
    headers={
        "X-Atlaso-Project-Key": ATLASO_KEY,
        "X-Atlaso-Subject": "user_8213",
    },
    json={"text": "Prefers window seats.", "tags": ["prefs"]},
).raise_for_status()

await fetch("https://mcp.atlaso.ai/developer/v1/memories", {
  method: "POST",
  headers: {
    "X-Atlaso-Project-Key": process.env.ATLASO_KEY,
    "X-Atlaso-Subject": "user_8213",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({ text: "Prefers window seats.", tags: ["prefs"] }),
});

Response

{ "id": "mem_9f2c…", "subject": "user_8213", "redacted": [] }

redacted lists the category names of any recognized secrets, credentials, or high-entropy tokens scrubbed before storing, such as aws_access_key or high_entropy. It never contains the removed values.

Recall

Retrieve the most relevant memories for a subject, semantically ranked. GET /recall?q=<query>&limit=<n>

Param	Notes
`q`	Required. The query, ≤2048 chars.
`limit`	Optional. Default 5; values are clamped to 1–50.

curl "https://mcp.atlaso.ai/developer/v1/recall?q=seating&limit=5" \
  -H "X-Atlaso-Project-Key: $ATLASO_KEY" \
  -H "X-Atlaso-Subject: user_8213"

import requests

r = requests.get(
    "https://mcp.atlaso.ai/developer/v1/recall",
    headers={"X-Atlaso-Project-Key": ATLASO_KEY, "X-Atlaso-Subject": "user_8213"},
    params={"q": "seating", "limit": 5},
)
r.raise_for_status()
print(r.json()["results"])

const url = new URL("https://mcp.atlaso.ai/developer/v1/recall");
url.search = new URLSearchParams({ q: "seating", limit: "5" });
const res = await fetch(url, {
  headers: {
    "X-Atlaso-Project-Key": process.env.ATLASO_KEY,
    "X-Atlaso-Subject": "user_8213",
  },
});
const { results } = await res.json();

Response

{
  "subject": "user_8213",
  "is_confident": true,
  "has_disagreement": false,
  "results": [
    {
      "id": "mem_9f2c…",
      "content": "Prefers window seats.",
      "tags": ["prefs"],
      "is_confident": true,
      "has_disagreement": false,
      "conflict_peers": []
    }
  ]
}

is_confident — whether Atlaso is confident in the result (top level = the overall answer; per-result = that memory).
has_disagreement / conflict_peers — set when the subject holds conflicting memories (e.g. “likes window seats” vs a later “now prefers aisle”). Atlaso surfaces both rather than silently picking one; you decide which to trust.

Delete a memory

Hard-delete one memory. Idempotent — deleting something already gone returns deleted: false, not an error. DELETE /memories/{id}?reason=<optional>

curl -X DELETE "https://mcp.atlaso.ai/developer/v1/memories/mem_9f2c…?reason=user_request" \
  -H "X-Atlaso-Project-Key: $ATLASO_KEY" \
  -H "X-Atlaso-Subject: user_8213"

{ "deleted": true, "id": "mem_9f2c…", "subject": "user_8213" }

This is a real erase — the memory, its search index, and its conflict links are removed, not tombstoned. Purge an end-user’s entire memory bag in one call — for a “delete my data” request. DELETE /subjects/{subject}?reason=<optional>

curl -X DELETE "https://mcp.atlaso.ai/developer/v1/subjects/user_8213?reason=gdpr" \
  -H "X-Atlaso-Project-Key: $ATLASO_KEY" \
  -H "X-Atlaso-Subject: user_8213"

The subject in the path must match the subject you’re authenticated as (mismatch → 403). Idempotent, and it frees the storage the subject used.

{ "purged": true, "subject": "user_8213", "erased": 12 }

Device credentials

Mint, batch-enroll, and revoke per-device credentials. These use the project key only (no subject header). Full guide: Devices & hardware.

Method	Path	Purpose
`POST`	`/device-credentials`	Mint one credential: `{ subject, external_device_id?, label?, group_id? }`.
`POST`	`/device-credentials/batch`	Enroll up to 500 devices at once (`Idempotency-Key` supported).
`POST`	`/device-credentials/revoke`	Revoke by `credential_id`, or a whole fleet by `group_id`.

curl -X POST https://mcp.atlaso.ai/developer/v1/device-credentials \
  -H "X-Atlaso-Project-Key: $ATLASO_KEY" \
  -H "Content-Type: application/json" \
  -d '{"subject": "kiosk_lobby", "label": "Lobby kiosk", "group_id": "lobby-fleet"}'

{
  "credential_id": "…",
  "full_token": "atldev_…_…",
  "subject": "kiosk_lobby",
  "external_device_id": null,
  "existing": false
}

full_token is shown once for a newly minted credential — flash it to the device and store nothing else. For batch enrollment, send the devices and an optional fleet-level group_id:

{
  "devices": [
    {
      "subject": "kiosk_lobby",
      "external_device_id": "kiosk-001",
      "label": "Lobby kiosk"
    }
  ],
  "group_id": "lobby-fleet"
}

The response wraps each credential in credentials and includes the total:

{
  "credentials": [
    {
      "credential_id": "…",
      "full_token": "atldev_…_…",
      "subject": "kiosk_lobby",
      "external_device_id": "kiosk-001",
      "existing": false
    }
  ],
  "count": 1
}

Re-enrolling the same external_device_id returns the existing credential with existing: true and full_token: null. Revocation returns the number of credentials revoked:

{ "revoked": 1 }

Get started

Tools

Build — developer API

Account

Help

Remember

Recall

Delete a memory

Device credentials

​Remember

​Recall

​Delete a memory

​Delete a subject (GDPR)

​Device credentials

Remember

Recall

Delete a memory

Delete a subject (GDPR)

Device credentials