Skip to content

MCP API — client setup

Every recipe below points an MCP client at https://api.kindo.ai/v1/mcp and authenticates with your Kindo API key in the Authorization header. Set KINDO_API_KEY in your shell, or substitute the literal value.

Terminal window
claude mcp add kindo --transport http https://api.kindo.ai/v1/mcp \
--header "Authorization: Bearer $KINDO_API_KEY"

For one-off agents, evaluations, or scripts, isolate from your global MCP config with a one-off file:

kindo-mcp.json
{
"mcpServers": {
"kindo": {
"type": "http",
"url": "https://api.kindo.ai/v1/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}
Terminal window
claude --strict-mcp-config --mcp-config ./kindo-mcp.json \
-p "List my open Linear tickets."

--strict-mcp-config ignores user/project MCP configuration and uses only the file you pass.

Other clients

Any spec-conformant MCP client that supports Streamable HTTP will work. Point it at https://api.kindo.ai/v1/mcp and send Authorization: Bearer YOUR_API_KEY. If your client speaks only stdio, wrap the URL with mcp-remote using the same pattern shown for Claude Desktop and Codex CLI.

Tool surface too large?

The bare endpoint shows every integration your account can reach (see Scoping). For broad-access accounts the catalog can exceed a model’s context window. Narrow it with a query filter:

https://api.kindo.ai/v1/mcp?integrations=linear,github

The filter can only narrow what your account already admits — it never grants new access. See kindo-extensions for the header form.

Next steps

  • Request shape — what to send, what comes back.
  • Scoping — which tools appear and why.
  • Errors — auth and protocol error envelopes.