evyAI API

The evyAI API lets a script, a tool or an AI agent act on one user’s evyAI account with a personal access token: read and write connections, notes, to-dos, personas, styles, assets and Agent memory, generate content and images in that user’s voice, create campaign drafts, publish to LinkedIn, and email the user. It shipped in October 2026 under API and MCP in Settings.

If you want Claude to use evyAI rather than writing code, you want the MCP connector instead. Same tools, no code.

Get a token

1. Sign in to evyAI and open Settings > API and MCP (direct link).
2. Click New token, tick the permissions it may use, and pick a lifetime (default 90 days).
3. Copy it. The token is shown once.

Send it on every call:

Authorization: Bearer <your token>

A token acts as the user who made it, limited to the permissions it was given. There is no organization-level key. Revoke a token from the same Settings page.

Shape of the API

The API is RPC-style, not REST. Every operation is a POST to the capability’s name with a JSON body:

POST https://services.api.evyai.com/api/capabilities/<capability.name>
Content-Type: application/json

{ ...input }

• GET https://services.api.evyai.com/api/capabilities returns the catalog: every capability’s name, description, scope (the permission it needs), effect (read, write or spend) and its input JSON Schema. An agent should call this first and discover the rest. It is the only GET.
• An OpenAPI 3.1 spec is public at https://services.api.evyai.com/api/capabilities/openapi.json (no token needed).
• Success: {"success": true, "data": {...}}. Failure: {"success": false, "errorId": N, "message": "..."}.
• Lists come back as {"total": N, "<plural>": [...]}. Every object carries its id and a deep-link url into the app.
• Dates are yyyy-MM-dd. Times are wall-clock in the timezone you send (an IANA name such as America/New_York).

Capabilities

37 capabilities, named <area>.<verb>:

Area	Capabilities	Effect
Profile and brand	profile.get, profile.update, brand.get, socialCard.get	read, write
Content	content.generate (post, ideas, comment, message and more, in the user’s voice)	spend
Images	image.generate (can use the user’s photos and logo)	spend
Email	email.sendToMe (only to the account’s own address)	write
Connections	connections.find, connections.get, connections.create, connections.update, connections.addToList, notes.add	read, write
To-dos	todos.list, todos.add, todos.update, todos.complete, todos.delete	read, write
Campaigns	campaigns.list, campaigns.createDraft, campaigns.createManual, campaigns.addPosts	read, write
LinkedIn	linkedin.publish (now or scheduled; needs LinkedIn connected in evyAI)	write
Personas and styles	personas.list, personas.create, personas.update, styles.list, styles.create, styles.update	read, write
Assets	assets.list, assets.get, assets.create, assets.update	read, write
Agent memory	memory.list, memory.add, memory.update, memory.delete	read, write

Prospect search, profile preview and unlocking people are available through the MCP connector and are not in the REST catalog at the time of writing.

Permissions

Each token carries the permissions you ticked. A call without the needed permission returns 403 and names the permission in message.

Permission	Covers
Read / change your own profile	profile.*, brand.get
Read your social card	socialCard.get
Read / add and change connections, notes and to-dos	connections.*, notes.add, todos.*
Write posts and other content (can cost coins)	content.generate, image.generate
Read / create campaigns	campaigns.*
Publish and schedule posts on your LinkedIn	linkedin.publish
Email things to your own address	email.sendToMe
Personas and writing styles	personas.*, styles.*
Assets	assets.*
What the Agent remembers about you	memory.*

What it costs

Effect	Capabilities	Charge
read	every *.get, *.list, *.find	free
write	create, update, add, complete, delete, email, campaigns, publish	free
spend	content.generate, image.generate	the same as the action in the app: 1 evyCoin minimum per text generation (more for longer output) and 5 evyCoins per image on the free plan; text is unlimited on Premium and Pro

Nothing else is metered. See evyCoins.

Rate limit

300 calls per minute per user, announced in standard IETF headers on every response:

ratelimit: "300-in-1min"; r=297; t=15
ratelimit-policy: "300-in-1min"; q=300; w=60

r is calls remaining, t is seconds until reset. Over the limit returns HTTP 429. Read r and back off instead of retry-looping.

Errors

HTTP	When
400	input fails validation (required field missing, value out of range)
401	missing, invalid, expired or revoked token
403	token lacks the permission; message names it
404	unknown capability, or an id that is not in this account
409	already exists (for example a duplicate LinkedIn URL on connections.create) or the memory is full
429	rate limit; wait a minute
500	the server could not finish; also returned for malformed JSON bodies

Gotchas we measured

1. Unknown fields are silently ignored. The schemas say additionalProperties: false, but a misspelled field name does not error, it just does nothing. Required and range checks do fire.
2. No pagination. connections.find returns at most 20 results and ignores offset. With a search term, total also tops out at 20. Narrow the search instead of paging. Other lists return up to 20 as well.
3. connections.find does not search LinkedIn URLs. It matches name, company, headline, email and phone. Search the person’s name.
4. Always send timezone. Leave it out and times are read as UTC.
5. image.generate returns a signed URL that expires in one hour. Download it right away. The image stays in the user’s in-app history.
6. content.generate with kind: "ideas" returns one string holding all the ideas, not one element per idea. Split it yourself.
7. No delete for most records. Connections, notes, personas, styles, assets and campaigns can be created and updated but not deleted through the API. Only to-dos and memories have a delete. Test writes leave real records; exercise the write path with a memory add, update and delete.
8. linkedin.publish is irreversible. Give date and time to schedule, omit both to post now. It fails with a pointer to Settings if no LinkedIn account is connected.
9. Campaigns are called workflows in parts of the response. Draft campaigns created through the API are reviewed and launched in the app.
10. Memory entries are 250 characters and the same ones the in-app Agent reads. What you write here changes how evyAI writes for that user.

Example

Read the user’s profile, then write a post in their voice:

curl -s -X POST https://services.api.evyai.com/api/capabilities/profile.get \
  -H "Authorization: Bearer $EVYAI_TOKEN" \
  -H "Content-Type: application/json" -d '{}'

curl -s -X POST https://services.api.evyai.com/api/capabilities/content.generate \
  -H "Authorization: Bearer $EVYAI_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"kind":"post","instructions":"Why follow up beats cold outreach. Include one story from my own experience and end with a question."}'

Check the input schema in the catalog for the exact fields each capability accepts; the catalog is the source of truth and may grow.

Related

• Connect evyAI to Claude (MCP)
• MCP connector for builders
• Chat URL API for deep links that open the Agent with a question

💬 Ask evyAI about this