How to connect DeepSeek to LinkedIn, step by step

What you need

• A Finchreach account (sign in at finchreach.com) and the LinkedIn account you want DeepSeek to work on.
• An MCP client to run DeepSeek in: Cline, OpenCode or Roo Code.
• About five minutes.

Step 1: Create a Finchreach API key

1. Sign in at finchreach.com and open API keys.
2. Create a key and name it after the tool, for example “DeepSeek”.
3. Copy it right away: it starts with lk_ and is shown only once.

Step 2: Connect your LinkedIn account

1. Open Accounts and click to connect LinkedIn.
2. Follow the steps until the account shows as connected.

Approval mode is on by default: nothing the assistant drafts reaches LinkedIn until you approve it.

Step 3: Run DeepSeek in an MCP client and add Finchreach

DeepSeek doesn’t connect to MCP servers itself, so the client you run it in does. Not applicable: the MCP client you run DeepSeek in sends the Finchreach key.

1. chat.deepseek.com and the DeepSeek app have no MCP or connector settings, and the API offers function calling only.
2. Use DeepSeek models inside an MCP client that supports headers, such as Cline or OpenCode: set DeepSeek as the model provider there, then add Finchreach with that client’s guide.

Client guides with the exact config:

• Cline (connects directly)
• OpenCode (connects directly)
• Roo Code (connects directly)

Step 4: Test the connection

In the client you connected, start with a read:

Ask your assistant“List my connected LinkedIn accounts, then summarize my five latest notifications.”

You should see calls to linkedin_list_accounts and linkedin_get_notifications. Then try a write, which goes to your approval queue instead of LinkedIn:

Ask your assistant“Draft a short LinkedIn post saying I’m trying a new AI assistant workflow. Submit it for my approval.”

Step 5: Approve or reject it in Finchreach

1. Open Approvals. The draft is there with its exact text and the account it would go out from.
2. Approve it to publish, or reject it (reject the test post if you don’t want it live).
3. The assistant can check the result with linkedin_get_action: pending, approved, rejected, executed, failed or expired.

Troubleshooting

• Pick a DeepSeek model that supports tool calling in your client; the model decides when to call the LinkedIn tools.
• 401 or “unauthorized”: the key is wrong, revoked or missing its prefix. Finchreach accepts Authorization: Bearer lk_… or X-API-Key: lk_…; check for a missing Bearer , quotes or a trailing space.
• A tool is missing: the key’s permissions don’t include it. Create a key with the permissions you need under API keys.
• “No LinkedIn account”: connect one under Accounts, or reconnect it if its session expired.
• A post or message “didn’t go out”: it is probably still waiting under Approvals. Writes run only after you approve them.

Still stuck? Compare your setup with DeepSeek’s documentation, checked on 2026-10-28; client screens change often.

What to ask it next

Ask your assistant“Summarize my LinkedIn feed in Chinese and English.”

Ask your assistant“Draft a LinkedIn post about our open-source release, technical but readable.”

Ask your assistant“List my unread LinkedIn conversations and draft replies.”

More ideas on the DeepSeek integration page.

Common questions