LinkedIn MCP Server

Q: What is Connector Builder?

Define exactly the inputs and outputs your agent needs. Pagination, joins, field mapping — all server-side. One action in, one response out. Test in the AI Playground before deploying.

Q: How does Code Mode work?

Agent gets search_tools and execute_code. It writes TypeScript in a sandboxed runtime. Raw responses (55K+ chars) stay in the sandbox. Only a 500-token summary returns to the LLM.

Q: How does StackOne handle LinkedIn authentication?

Full OAuth lifecycle managed automatically. Each end user authenticates their own LinkedIn account. Tokens stored, refreshed, rotated. Your infrastructure never touches credentials.

Q: Does StackOne store my LinkedIn data?

No. Realtime proxy. Data flows through, never persisted. Audit logs record actions and actors — not payloads.

Q: Can I customize which LinkedIn actions my agent can access?

Yes. Connector Builder scopes actions to your use case. Glob-pattern filtering narrows thousands of actions to the handful your agent needs.

Q: What frameworks does StackOne MCP work with?

Any MCP client: Claude Desktop, Claude Code, Goose. Framework SDKs: OpenAI, Anthropic, Vercel AI, Google ADK, CrewAI, Pydantic AI, LangChain. Also REST API, AI Action SDK, and A2A.

Enhance LinkedIn tool access, agentic execution, and security all at once with StackOne managed MCP server.

Start Free Book Demo Read Docs

pre-built from LinkedIn's enterprise API

96%

context reduction with Code Mode

90.8%

prompt injection detection accuracy

Tool Access

2+ Actions. Full API Coverage.

Built on LinkedIn's enterprise API
Dynamic tool generation per integration
Natural language discovery via Tool Search

Extensibility

Custom Actions Scoped to Your Use Case.

Define exact input/output schemas
Server-side joins, pagination, field mapping
One action in, one response out
Test in the AI Playground before deploying

Explore Connector Builder

Authentication

Per-User OAuth. Real Multi-Tenancy.

Each user authenticates their own LinkedIn account
Automatic token refresh and rotation
Per-tenant credentials, permissions, audit logs
Multiple linked accounts per user
Zero credential storage on your infra

Learn about Managed Auth

Security

90.8% Detection. Zero Data Stored.

Two-tier pipeline: pattern matching <1ms, ML ~4ms
Scans every tool response automatically
Realtime proxy — data flows through, never stored
Audit logs record actions, not payloads

See Defender benchmarks

Fewer Calls. Less Context.
Faster Agents.

Agentic Execution

3 Tool Calls Instead of 15.

Connector Builder composes multiple API calls server-side
Each eliminated call saves a full model reasoning cycle
Custom I/O schemas reduce token bloat per call
Test composed actions in the AI Playground

Connector Builder docs

connector-builder.ts

// Without Connector Builder: 15 tool calls
agent.call("list_employees");
agent.call("get_employee", { id });
agent.call("get_department", { id: emp.deptId });
// ... 12 more round trips

// With Connector Builder: 1 composed action
agent.call("get_employee_with_context", { id });
// Returns employee + dept + manager

96% fewer tokens 5x fewer tool calls

Context Optimization

2 Tools. 96% Less Context.

Agent gets search_tools and execute_code
Writes TypeScript in a sandboxed runtime
Raw API responses (55K+ chars) stay in the sandbox
Only a 500-token summary returns to the LLM

When to use Code Mode: workflows that touch 2+ providers, cross-reference data, or return responses larger than 10K tokens.

Read about Code Mode

AI Client

Claude / Cursor

Code-Mode

Gateway

Credentials injected
at transport layer

GitHub, Slack

Custom MCP Servers

Every LinkedIn Tool Call
Scanned Automatically.

Built into every managed connector. No setup. No configuration. Two-tier defense pipeline catches known patterns in ~1ms, then runs ML classification in ~4ms for adversarial attacks.

Defender

0.91

F1 Score

22 MB ~10ms

LLM Moderation

0.82

F1 Score

>2 GB 500ms+

Gateway Proxy

0.65

F1 Score

Infra required

Regex Patterns

0.40

F1 Score

<1 MB

Two-Tier Defense Pipeline

Tier 1 catches known attack patterns in ~1ms. Tier 2 runs a fine-tuned ML classifier in ~10ms for adversarial attacks that evade pattern matching.

Scans Every Tool Response

Emails, documents, PRs, CRM records — any tool response can contain hidden instructions. Defender scans and sanitizes content before your agent processes it.

Open Source, Zero Latency

Runs in-process with a bundled ONNX model — no external API calls, no inference costs, no network latency.

Explore Defender 10 Real Attack Examples

Connect to LinkedIn in 3 Steps

Activate & Customize

Activate LinkedIn and customize actions with Connector Builder.

Link Accounts

Link accounts — per-user OAuth handled by StackOne.

Connect Any Client

Connect any MCP client to your StackOne endpoint.

One Endpoint.
Every Framework and App.

https://api.stackone.com/mcp

Frameworks

Apps & IDEs

Claude Desktop config

claude_desktop_config.json

{
  "mcpServers": {
    "stackone": {
      "url": "https://api.stackone.com/mcp?x-account-id=<id>",
      "headers": { "Authorization": "Basic <token>" }
    }
  }
}

Quickstart guide

Not Just MCP

REST API

Direct API access for any backend.

AI Action SDK

Native tool calling without MCP.

A2A

Google's agent-to-agent protocol.

Common Questions

What is Connector Builder?

Define exactly the inputs and outputs your agent needs. Pagination, joins, field mapping — all server-side. One action in, one response out. Test in the AI Playground before deploying.

How does Code Mode work?

Agent gets search_tools and execute_code. It writes TypeScript in a sandboxed runtime. Raw responses (55K+ chars) stay in the sandbox. Only a 500-token summary returns to the LLM.

How does StackOne handle LinkedIn authentication?

Full OAuth lifecycle managed automatically. Each end user authenticates their own LinkedIn account. Tokens stored, refreshed, rotated. Your infrastructure never touches credentials.

Does StackOne store my LinkedIn data?

No. Realtime proxy. Data flows through, never persisted. Audit logs record actions and actors — not payloads.

Can I customize which LinkedIn actions my agent can access?

Yes. Connector Builder scopes actions to your use case. Glob-pattern filtering narrows thousands of actions to the handful your agent needs.

What frameworks does StackOne MCP work with?

Any MCP client: Claude Desktop, Claude Code, Goose. Framework SDKs: OpenAI, Anthropic, Vercel AI, Google ADK, CrewAI, Pydantic AI, LangChain. Also REST API, AI Action SDK, and A2A.

Deep Dives

Agentic Context Engineering: Why AI Agents Kill Their Own Context Windows

AI agents exceed their context windows without knowing it. Six failure patterns and seven survival architectures for agentic context engineering.

15 min

Your Agent Toolkit: MCP Server, SDK-based Toolset, or Both?

MCP promises standardization but framework support varies wildly. SDKs give control but mean vendor lock-in. The hybrid approach fills gaps that pure options leave open.

9 min