Do I still need to paste an API key into Cursor or Claude?

No. With remote MCP OAuth, the client opens Gluedly login, you approve access, and Gluedly issues a dedicated MCP API key as the access token.

Does this replace the Gluedly MCP stdio package?

No. Remote OAuth is the preferred path for hosted agents. The stdio @gluedly/mcp-server package remains supported when you want a local bridge.

What happens to my existing API keys?

Manual API keys keep working for REST and other integrations. OAuth consent rotates a dedicated MCP-sourced key so reconnects stay scoped to MCP.

Which clients does Gluedly MCP OAuth work with?

Any MCP client that supports remote HTTP MCP plus OAuth 2.1 with PKCE — including Cursor, Claude, Codex, and similar agent hosts.

October 1, 2026 · Product updates · Donatas · 2 min read

Connect Cursor, Claude, and Codex to Gluedly MCP Without Pasting API Keys

One-click MCP auth for AI agents

Gluedly now hosts remote MCP with OAuth 2.1. When Cursor, Claude, Codex, or any MCP client hits a 401, it opens Gluedly login, you approve access, and we mint a fresh MCP API key — no copy-paste, no stale tokens.

What changed?

Remote MCP URL — point your client at /mcp instead of only local stdio.
OAuth consent — login + approve regenerates a dedicated MCP key.
stdio still works — keep @gluedly/mcp-server when you want a local bridge.

#Automation #API #Agents #MCP #OAuth #Cursor #Claude

Connect Cursor, Claude, and Codex to Gluedly MCP Without Pasting API Keys

AI agents are only useful if they can reach your scraping workspace without friction. Until now, wiring Gluedly into Cursor, Claude, Codex, or similar MCP clients usually meant creating an API key, pasting it into a config file, and hoping it never expired or leaked.

That flow breaks the moment a token is wrong, revoked, or shared across machines. Agents get a silent 401. You dig through docs. Momentum dies.

Remote MCP with OAuth, built for real clients

Gluedly now speaks the same language modern MCP hosts expect: Streamable HTTP MCP plus OAuth 2.1 (PKCE, protected-resource metadata, and dynamic client registration).

Point your client at your Gluedly MCP URL. On the first unauthorized request the agent is challenged to authenticate. You sign in to Gluedly, approve the client, and we rotate a dedicated MCP API key for that connection. The access token the client stores is that key — ready for tools like listing pages and reading scrape data.

No pasted secrets in mcp.json. No “which key was this again?” support thread. Reconnects stay as simple as approving again when you need a fresh credential.

What you can do today

Cursor / Claude / Codex and other OAuth-capable MCP clients — use the remote MCP endpoint and complete the browser consent flow.
Dedicated MCP keys — OAuth-issued keys are tracked separately from manual keys, so rotation on consent does not surprise your other integrations.
Keep stdio when you want it — @gluedly/mcp-server remains available for local or scripted setups; remote OAuth is the preferred path for hosted agents.

Why this matters for scraping teams

Agents that can list pages, inspect mappings, and pull scrape results only help if auth is boring. OAuth makes Gluedly feel like every other SaaS your team already trusts: redirect, approve, continue working.

That is the difference between “we tried MCP once” and “the agent is part of the daily scrape workflow.”

Open Gluedly and connect MCP

Prefer the details?

Remote MCP configuration, OAuth endpoints, and stdio fallback notes live in your account API docs and integration guides after you sign in. Start with the remote URL, let the client discover auth, and approve once — then get back to shipping scrapes.