Connect an AI agent
Help an agent discover Mjhr safely before it reads workspace data.
You will know which context an agent should read and what it must confirm before acting.
What an agent needs
An agent should not guess how Mjhr works. It should read the product help, load the API contract, ask for a scoped credential when needed, and confirm the workspace or saved search before using data.
The help package exposes human-readable pages and machine-readable resources so agents can start with the same source of truth as users and developers.
Discovery order
- 1
Read llms.txt
Start at /llms.txt for the concise product map and links to deeper context.
- 2
Fetch OpenAPI
Use /openapi.json as the canonical HTTP contract before constructing requests.
- 3
Use MCP when available
Call /mcp when your runtime supports MCP tools. The tools mirror OpenAPI operations and delegate to the existing API.
- 4
Ask for missing credentials
Do not invent API keys. Ask the user to provide a scoped key or connect a secret store.
- 5
Confirm before using data
Show the workspace and saved search that will be queried before using results for analysis, outreach, or export workflows.
MCP request shape
The MCP endpoint speaks JSON-RPC over HTTP. Send Authorization on the MCP request; tool calls forward that header to the API service.
{
"jsonrpc": "2.0",
"id": "tools",
"method": "tools/list"
}{
"jsonrpc": "2.0",
"id": "searches",
"method": "tools/call",
"params": {
"name": "list_searches",
"arguments": {
"query": { "limit": 25 }
}
}
}Agent safety boundaries
- Treat API keys as secrets and avoid writing them into generated files or logs.
- Prefer read-only scopes unless the user explicitly asks the agent to create folders.
- Explain which workspace and saved search will be queried before using results.
- Use public IDs returned by the API, not internal database IDs.