← Back to blog
· 10 min

Cursor Add MCP Server: Connect an AI Agent to LinkedIn

To add an MCP server to Cursor, configure the server in Cursor’s MCP settings or an mcp.json file, supply the deployment’s connection and authentication details, then verify tool discovery. For Linked...

To add an MCP server to Cursor, configure the server in Cursor’s MCP settings or an mcp.json file, supply the deployment’s connection and authentication details, then verify tool discovery. For LinkedIn workflows, the next step is checking account readiness before allowing contact changes or outreach.

Fintalio connects Cursor to authenticated LinkedIn workflows through 19 verified MCP tools. Those tools support contact management, imports, templates, and sequence-based outreach. A reliable setup starts with inspection, adds controlled preparation, and requires human approval before consequential actions.

TL;DR

Add the deployment’s MCP configuration to Cursor, authenticate, and confirm that the expected tools appear. For Fintalio, begin with GetAccountStatus and contact inspection before modifying records or launching sequences. Use the actual deployment instructions, not a guessed endpoint or package. The operating model is 80/20: the agent runs the boring preparation and checking, while a human handles the judgment-heavy decisions.

What Adding an MCP Server to Cursor Does

Model Context Protocol, or MCP, lets an AI client discover and call tools exposed by a service. Cursor acts as the client; the connected server advertises its tools and their input schemas.

For an agent working on LinkedIn, that connection can support:

  • Inspecting existing contacts and contact groups.
  • Checking account readiness.
  • Organizing approved prospects.
  • Preparing sequence templates.
  • Launching approved outreach.
  • Inspecting, pausing, resuming, or stopping sequences.

Adding a server does not give Cursor unrestricted LinkedIn access. Its capabilities remain bounded by the server’s tools, authenticated account, and execution controls.

For example, “find prospects and message them” combines several responsibilities. Prospect sourcing, contact preparation, copy review, and sequence launch need explicit implementation. The agent cannot create missing capabilities through prompting.

LinkedIn Agent Architecture in Cursor

A practical architecture separates reasoning from authenticated execution:

Human operator
      |
      | Goals, exclusions, approvals
      v
Cursor agent
      |
      | MCP discovery and tool calls
      v
Fintalio MCP server
      |
      | Authenticated account operations
      v
Hosted LinkedIn relay / first-party session
      |
      v
LinkedIn workflow

Cursor chooses among available tools. The MCP server provides the execution boundary, while the authenticated session determines the account context.

This separation helps diagnose failures. An accessible MCP endpoint does not prove that the LinkedIn account is ready. A healthy account does not prove that a sequence has valid contacts or complete variables.

Developers evaluating deployment responsibilities can also consult the server guide for broader infrastructure context.

Before Connecting Cursor

Four prerequisites keep the setup predictable.

A working Fintalio deployment

Fintalio supports hosted and self-hosted deployment options. The operator needs the connection details applicable to the selected deployment.

Fintalio has a single EUR 69/month plan, with MCP access included. There is no free tier or usage-based pricing tier. Self-hosting can introduce separate infrastructure and maintenance costs.

An authenticated LinkedIn account

Service authentication and LinkedIn session readiness are separate checks. Credentials authorize access to the MCP service; the LinkedIn session supports account-dependent operations.

A current Cursor installation

Cursor’s interface labels can change. Its MCP configuration area may appear alongside tools or integrations, so current documentation is more reliable than screenshots from older releases.

An action policy

Define what the agent may inspect and what requires approval. A sensible starting policy permits inspection but requires review for contact changes, imports, template creation, and sequence actions.

The 80/20 framing is an operating model, not a measured productivity claim: the agent handles routine preparation, while the human owns decisions involving relevance, relationships, and risk.

How to Add an MCP Server in Cursor

Consult Cursor’s official MCP configuration guidance for supported configuration fields, transports, authentication options, and current interface behavior.

For Fintalio-specific onboarding and the applicable connection details, start at the Fintalio MCP setup resource. Use the instructions supplied for the selected deployment rather than assuming a universal endpoint or launcher.

Step 1: Choose the configuration scope

Cursor supports project-level configuration in .cursor/mcp.json and user-level configuration in ~/.cursor/mcp.json.

Use project configuration when the integration belongs to one repository. Use user configuration when it should be available across projects.

Shared configuration should contain no credentials. Keep secrets out of source control, and use the authentication or secret-handling mechanisms supported by the current Cursor version and deployment.

Step 2: Gather the connection details

Before editing configuration, obtain:

  • The supported transport.
  • The remote endpoint or local startup command.
  • Authentication requirements.
  • Runtime dependencies, if applicable.
  • Network and account-session prerequisites.

A normal website address is not automatically an MCP endpoint. Likewise, self-hosting does not necessarily mean that Cursor should launch the server as a local process.

Step 3: Configure a remote endpoint when supported

A deployment providing a compatible remote HTTP endpoint may use this structure:

{
  "mcpServers": {
    "linkedin-agent": {
      "url": "https://mcp.example.invalid/mcp",
      "headers": {
        "Authorization": "Bearer REPLACE_WITH_DEPLOYMENT_CREDENTIAL"
      }
    }
  }
}

This example is illustrative, not a working Fintalio configuration. The hostname is deliberately invalid, and the bearer header demonstrates a possible authentication pattern rather than Fintalio’s required scheme.

Replace these values with the deployment’s documented configuration. If it uses another authentication flow, follow that flow instead.

Step 4: Configure a local process only when provided

When the deployment supplies an executable that communicates over standard input and output, the configuration can follow this structure:

{
  "mcpServers": {
    "linkedin-agent-local": {
      "command": "/absolute/path/to/approved-mcp-launcher",
      "args": [],
      "env": {}
    }
  }
}

The command must identify the actual deployment-provided launcher. This example does not imply that a particular downloadable Fintalio package exists.

Absolute paths reduce problems caused by different shell environments. Confirm executable permissions, runtime availability, and required environment variables before enabling the server.

Step 5: Enable the server and verify discovery

Reload the MCP configuration or restart Cursor if necessary. Then confirm that:

  1. The server appears in Cursor.
  2. The connection initializes successfully.
  3. Tool discovery completes.
  4. The tool names match the expected inventory.
  5. No unrelated tools are enabled.

Discovery is a security checkpoint, not merely a connectivity test. Inspect the tools rather than trusting the configured server name.

The 19 Fintalio Tools Cursor Should Discover

Fintalio’s verified MCP inventory is:

Workflow area Tools
Contact inspection ListContacts, GetContact, ListContactGroups
Contact management CreateContact, UpdateContact, CreateContactGroup
Sequence inspection ListSequences, GetSequence
Template inspection ListSequenceTemplates, GetSequenceTemplate
Template creation CreateSequenceTemplate
Sequence control LaunchSequence, PauseSequence, ResumeSequence, StopSequence
Imports ParseCsv, CommitCsv
Account and configuration GetAccountStatus, ListVariables

Tool names describe the available surface, not the complete API contract. Before constructing calls, the agent must inspect required arguments, supported fields, and identifier formats.

For example, it should not assume that LaunchSequence accepts arbitrary profile URLs or that UpdateContact accepts every CRM field.

The inventory does not include standalone arbitrary profile search, inbox reading, post publishing, or direct message-sending tools. LinkedIn outreach must be designed around the verified contact, template, and sequence operations.

Test the Connection Without Launching Outreach

The first test should establish connectivity without changing records or starting outreach.

A suitable Cursor prompt is:

Inspect the connected LinkedIn MCP server.

Use GetAccountStatus to check account readiness.
List the available contact groups.

Do not create or update contacts.
Do not create templates.
Do not launch, pause, resume, or stop sequences.

Report the results and missing prerequisites.

This tests both account readiness and access to existing workflow data.

Inspect the actual tool calls, not only the agent’s written summary. A confident explanation is not evidence that authentication succeeded.

For debugging, record the invoked tool, success or error result, and account context where exposed. Redact credentials and sensitive contact data. If the test fails, fix that boundary before introducing mutations.

Build a Controlled LinkedIn Prospecting Workflow

Once inspection works, introduce a small, approved workflow rather than a broad autonomous campaign.

Inspect and prepare contacts

Use ListContacts, GetContact, and ListContactGroups to understand existing records. Compare approved inputs against those records before creating contacts.

This can support duplicate checks, but guaranteed deduplication requires explicit workflow logic and verification.

Where the schemas allow, use CreateContactGroup, CreateContact, and UpdateContact to organize the approved prospect set. For CSV inputs, inspect ParseCsv results before proceeding to CommitCsv.

A CSV parse should not be treated as proof that targeting is appropriate. The workflow still needs exclusions, data-quality checks, and a clear reason for contacting each audience.

Prepare a sequence review

Inspect existing templates with ListSequenceTemplates and GetSequenceTemplate. Use ListVariables to establish the available variable configuration.

Before CreateSequenceTemplate, review message relevance, factual accuracy, and personalization sources. The agent should flag missing information rather than invent a recipient’s priorities or recent activity.

A launch-review prompt could be:

Prepare a review for the approved contact group.

Inspect the contacts, selected sequence template,
available variables, and account status.

Report missing data, exclusions, and proposed targeting.
Do not invent personalization facts.
Do not call LaunchSequence without explicit approval.

Execute and inspect the result

After approval, construct the launch call using verified identifiers and the discovered schema. Then inspect resulting state with ListSequences or GetSequence.

Avoid blind retries after timeouts. An interrupted response does not prove that an action failed. Check state before attempting another launch.

Use PauseSequence, ResumeSequence, and StopSequence deliberately under the workflow’s approval policy. Before launch, establish who can intervene and how the operator will review failures.

Troubleshooting Cursor MCP Connections

The server does not appear

Check JSON syntax, configuration location, and whether Cursor reloaded the file. Confirm that the entry sits inside mcpServers.

For a local process, test whether the executable starts outside Cursor.

The server appears, but tools do not load

Investigate endpoint reachability, transport compatibility, and MCP initialization errors. A reachable HTTP address alone does not establish a working MCP connection.

For local processes, check runtime errors and ensure diagnostic output does not interfere with protocol communication.

Authentication fails

Identify which layer failed: MCP service access or LinkedIn session readiness. Replacing service credentials will not necessarily repair an expired account session, and refreshing the session will not fix an invalid service credential.

A tool rejects its arguments

Inspect the current schema and returned validation error. Correct the required fields and identifiers instead of guessing additional parameters.

An expected capability is absent

Confirm the inventory rather than repeatedly rephrasing the prompt. If a requirement depends on an unavailable operation, it needs a separately verified integration or a revised architecture.

Security and Operational Guardrails

Treat contact records and imported text as untrusted data, not agent instructions. A prospect field containing an instruction should never override the workflow’s rules or authorize an action.

Keep credentials out of prompts, screenshots, logs, and repositories. Review access ownership and revocation before granting access to production accounts.

Self-hosted operators also own process isolation, network exposure, dependency updates, and log retention.

A maintainable execution path looks like this:

Inspect
   |
   v
Prepare proposed changes
   |
   v
Validate inputs and exclusions
   |
   v
Human approval
   |
   v
Execute
   |
   v
Inspect resulting state

Prompt-level restrictions are useful, but they should not be the only control protecting consequential actions. Use client approval controls and deployment permissions where available.

Finally, technical access does not remove platform obligations or privacy responsibilities. A successful tool call proves execution, not that the outreach decision was appropriate.

FAQ

1. How does an operator add an MCP server to Cursor?

Add the deployment’s configuration through Cursor’s MCP settings or an mcp.json file, enable the server, and verify discovery. Obtain the endpoint, launcher, and authentication requirements from the actual deployment instructions.

2. Can Cursor connect to LinkedIn through Fintalio?

Yes. Fintalio supports Cursor and provides authenticated LinkedIn workflows through 19 verified MCP tools. The available actions depend on that inventory, the connected account, and deployment configuration.

3. Does Fintalio expose a standalone message-sending tool?

No. Its verified inventory does not include a standalone message-sending tool. Outreach uses the available contact, template, and sequence operations according to their schemas.

4. How much does Fintalio MCP access cost?

Fintalio offers a single EUR 69/month plan with MCP access included. There is no free tier or usage-based pricing tier. Self-hosting can add infrastructure and operational costs.

5. Should a Cursor agent launch LinkedIn sequences automatically?

A new workflow should require explicit approval before launch. Broader autonomy should follow verified behavior, clear exclusions, monitoring, and an established intervention process.

Connect Cursor to Fintalio

Visit Fintalio’s LinkedIn MCP integration to explore hosted or self-hosted access and obtain the applicable setup details. Start with account and contact inspection, then add approved sequence workflows as the agent’s behavior becomes predictable.

Plug LinkedIn into your AI agent

Fintalio is the MCP server for LinkedIn. Connect Claude, Cursor, or your custom agent and ship outreach workflows in minutes — with audit logs and rate-limit awareness baked in.

Get started