An agent you already use can edit a workflow for you. It gets a key that works on that one workflow and nothing else: it can read the steps, change the draft, check it and test it. Publishing stays with you in Kaer.
Before you begin
- A workflow to change. The agent edits its draft; see Workflows.
- Claude Code, Cursor, Codex, or another agent that can connect to an MCP server over Streamable HTTP.
- A clear description of the change you want, as you would give a colleague.
Set it up
Create the key
Open the workflow, then Settings → Use it elsewhere → Create an editing key. The key is shown once.
Add it to your agent
Pick Claude Code, Cursor or Codex and copy the setup — or pick Any agent and paste the brief into the chat. The brief tells the agent how to connect and how to work.
Ask for the change
"Add a step that sends me the summary on Slack." The agent reads the workflow, saves the change, checks it and runs a test.
Review and publish
The change shows in the workflow's chat and editor. Publish when you are happy with it.
claude mcp add --transport http kaer-workflow-42 https://app.kaer.ai/api/mcp/workflows/42 \
--header "Authorization: Bearer kaer_sk_YOUR_KEY" \
--header "Accept: application/json, text/event-stream"For agents
If you are an agent setting this up yourself: the server speaks MCP over Streamable HTTP at https://app.kaer.ai/api/mcp/workflows/<id>. Send Authorization: Bearer <key> and Accept: application/json, text/event-stream. It offers six tools.
| Tool | What it does |
|---|---|
get_workflow | The steps and connections, with updated_at |
get_node_catalog | Every step kind, its settings and defaults |
save_workflow | Replaces the whole graph — pass expected_updated_at from get_workflow |
validate_workflow | Checks structure and settings, with no side effects |
simulate_workflow | A dry run with an input and a stand-in result per step |
estimate_hosting | Monthly credits for a number of runs |
1. get_workflow first. Change only what the owner asked for.
2. save_workflow replaces the whole graph — send every step back.
3. validate_workflow, then simulate_workflow, before you say you are done.
4. Never print the key. Publishing stays with the owner in Kaer.The full guide is also plain text for agents at https://kaer.ai/docs/llms-full.txt.
Common questions
Can the agent publish the workflow?
No. Publishing stays with you in Kaer. The agent can read, change, check and test the draft.
Can one key edit several workflows?
No. Each editing key is bound to one workflow. Create a key per workflow.
Does the agent's test run cost credits?
A simulated run uses stand-in results for each step, so nothing outside Kaer is sent or written. Live runs are the ones that bill credits. See Workflows.
How do I stop the agent's access?
Revoke the key in Settings → Developers. It stops working at once.