Give your AI agent customer context, with control.
Run Buddy MCP locally over stdio. Your AI agent can read subscribers, templates, messages and campaigns using a scoped Buddy credential. Writes stay off unless you enable them and approve each request in your terminal.
Read-only connection · your workspace
stdio
Your API key
By default
Read the data without copying it into chat.
Connect an MCP client that supports local stdio processes. Buddy's API enforces the credential's workspace, scopes and test or live mode.
Manual context
- Looking up delivery statuses by hand.
- Copying campaign figures into another tool.
- Checking templates before each integration change.
Scoped tools
- Read current message and campaign records.
- Inspect approved templates.
- Use test credentials without real sends.
Connect locally. Keep control of writes.
Use Node.js 18 or later and an MCP client with local stdio support. The beta package is available on npm; the repository README also documents local checkout setup.
Configure your client
Replace the key placeholder in your private client configuration. Keep credentials out of prompts and shared files.
Read your workspace
Ask for a message status, available templates or a campaign's results.
Approve any write
Writes require BUDDY_MCP_ALLOW_WRITES=true and approval in a controlling terminal. Desktop hosts without that terminal remain read-only. A model cannot approve its own request.
Know exactly which tools are available.
Every tool calls the same public API as your own integration.
Eight read tools, three optional writes.
Read workspace, subscriber, template, message and campaign data. Optional writes send one message, create a subscriber with consent evidence, or ingest an event.
Tool arguments never count as human approval.
Least privilege
Grant only the API scopes your connection needs.
Local transport
The client launches a stdio process. There is no hosted MCP URL or browser sign-in flow.
Safe retries
Reuse the same idempotency key when retrying a write with an uncertain outcome.
Useful context, explicit control.
Check status without switching tools.
Your agent reads the current API record and explains it in context. Customer-provided text remains untrusted data, never instructions.
Check a public message ID returned by Buddy.
Questions your agent can answer.
Delivery status
Check the status of this message ID.
SupportCampaign results
Read the results for this campaign ID.
MarketingTemplate checks
List the templates available in this workspace.
DevelopersThree things people ask about MCP server.
Can a model approve a send?
No. The local operator must approve the exact request in a separate terminal prompt. Missing or declined approval prevents the API call.
Questions about MCP server.
Anything else, ask Sarah.She reads every message and writes back herself.Which clients work?
Clients that can launch a local stdio MCP process. A client that only accepts hosted HTTP servers cannot use this transport.
Does this create or send campaigns?
No. Campaign tools are read-only. Use the Buddy dashboard or REST API for campaign creation and sending.
How do I revoke access?
Revoke the API key or OAuth connection in the developer dashboard, then remove the local client configuration.
Where do I put an OAuth token?
Set BUDDY_ACCESS_TOKEN in the process environment. Your host must refresh expiring tokens and restart the process with the replacement.
The rest of the Buddy developer platform.
Developers overviewConnect your agent with a test key.
Follow the public API quickstart, then connect a read-only local client.
No real sends
Scoped tools