mcp-server-linkedin
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TIMEOUT | No | Timeout for a single browser page operation, in milliseconds. Default is 5000. | 5000 |
| CHROME_PATH | No | Path to a Chrome/Chromium executable when Chrome is installed in a non-standard location. | |
| PROXY_BYPASS | No | Comma-separated hosts that bypass the proxy. Example: localhost,127.0.0.1,::1 | localhost,127.0.0.1,::1 |
| PROXY_SERVER | No | Proxy for browser traffic only, in scheme://host:port form. Supported schemes are http, https, socks4, and socks5. | |
| TOOL_TIMEOUT | No | Timeout for a whole tool call, in seconds. Default is 180. | 180 |
| PROXY_PASSWORD | No | Password for proxy authentication. | |
| PROXY_USERNAME | No | Username for proxy authentication. | |
| UV_HTTP_TIMEOUT | No | Timeout in seconds for uv's HTTP requests. The recommended config sets this to 300 to avoid slow first-run dependency downloads. | 300 |
| LOGIN_INLINE_WAIT | No | How long a tool call waits for a login to finish before telling the model to retry, in seconds. Default is 25, maximum is 45. | 25 |
| INSTALLER_TEMP_DIR | No | Existing directory to use for browser installation temporary files. | |
| LINKEDIN_MCP_CONTAINER | No | Overrides container detection. Set to false when running on a non-container host that is incorrectly detected as Docker, or true to force container behavior. | |
| AUTO_IMPORT_FROM_BROWSER | No | Controls whether the server imports a LinkedIn session from a signed-in local browser on the first tool call that needs one. Equivalent to --auto-import / --no-auto-import. | |
| FASTMCP_HTTP_ALLOWED_HOSTS | No | JSON array of additional Host header names the HTTP server should accept, for example '["mcp.example"]'. | |
| EXPERIMENTAL_PERSIST_DERIVED_RUNTIME | No | When set, keeps the derived browser runtime profile in Docker instead of rebuilding it from scratch on every start. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| logging | {} |
| prompts | {
"listChanged": true
} |
| resources | {
"subscribe": false,
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| get_person_profileC | Get a specific person's LinkedIn profile. |
| search_peopleC | Search for people on LinkedIn. |
| connect_with_personA | Send a LinkedIn connection request or accept an incoming one. The tool is annotated with destructiveHint so MCP clients will prompt for user confirmation before execution. |
| get_sidebar_profilesA | Get profile links from sidebar recommendation sections on a LinkedIn profile page. Extracts profiles from "More profiles for you", "Explore premium profiles", and "People you may know" sidebar sections. Follows "Show all" links to return the full list from each section. Sections that redirect to linkedin.com/premium are skipped. |
| get_my_profileA | Get the authenticated user's own LinkedIn profile. Navigates to /in/me/ and resolves the redirect to obtain the real username before reading the profile, so the url field in the result is the actual profile URL (e.g. linkedin.com/in/johndoe/) rather than /in/me/. |
| get_company_profileC | Get a specific company's LinkedIn profile. |
| get_company_postsC | Get recent posts from a company's LinkedIn feed. |
| search_companiesB | Search for companies on LinkedIn. |
| get_company_employeesA | List employees at a company from the LinkedIn /people/ page, including the demographics aggregate that this view exposes: where employees live, where they studied, and a function breakdown (Engineering, Sales, Operations, etc.). The demographics are unique to this tool. For filtered search by network degree (1st/2nd/3rd) or location, prefer search_people with current_company set to the company URN id. That path also returns more result pages than the /people/ tab. The optional keywords filter narrows results by name, title, or skill. company_name must be the exact LinkedIn URL slug (the path segment after /company/), not the display name. LinkedIn assigns unique slugs and the display name often does not match. For example, the AI lab Anthropic lives at /company/anthropicresearch/, not /company/anthropic/. If you are unsure of the slug, call search_companies first and pick the slug from the returned references. |
| get_job_detailsA | Get job details for a specific job posting on LinkedIn. |
| get_job_apply_urlA | Get how a job posting takes applications, and the employer's application link. Reads the posting without clicking anything. |
| search_jobsA | Search for jobs on LinkedIn. Returns job_ids that can be passed to get_job_details for full info. LinkedIn is moving accounts to a redesigned job search, which keeps only the keywords, date_posted and paging. It drops the filters marked "Classic search only" and reports them in section_errors as filters_dropped. When it drops the location or work type, search again with them in the keywords ("remote python developer in Lyon") and keep doing so; words such as "senior" or "internship" there narrow the results too. Quotes do not make a phrase exact on that search. |
| get_saved_jobsA | List the authenticated user's jobs at one stage of LinkedIn's job tracker. Returns job_ids that can be passed to get_job_details for full info. |
| get_inboxC | List recent conversations from the LinkedIn messaging inbox. The returned inbox text and result URL come from the ordinary messaging inbox. Click-derived conversation references are collected separately after requesting the compose page, which avoided inbox auto-opening on the measured variant. A row contributes a click-derived reference only after its click is followed by an observed different thread path. The scan stops at its first unverifiable click. section_errors.inbox reports that stop or unavailable scan rows; captured inbox text and independently extracted anchors retain their normal handling. A known thread_id can bypass row attribution when calling get_conversation. |
| get_conversationA | Read a specific messaging conversation. Provide either linkedin_username or thread_id to identify the conversation. When looked up by linkedin_username, resolution searches the messaging inbox for the participant's display name and click-visits every matching row to capture its thread ID — LinkedIn's sidebar has no anchor hrefs or thread-id attributes, so this is the only available path. Each visit selects the row in the LinkedIn UI and may mark it as read. Pass thread_id directly to skip this enumeration. Username resolution scans matching rows from a requested compose page first. Its indexable sequence ends before the first unresolved click, missing matching click target, or admitted row that fails the existing exact display-name check. An index outside that verified prefix is refused with its reason. Search is a fallback only when the inbox scan has no observed matching result and no such barrier; it never substitutes a result after a stopped or gapped inbox scan. Pass a known thread_id to bypass username/index resolution. |
| search_conversationsC | Search messages by keyword. Click-derived references require an observed different thread path after each row click. The first unverifiable click stops further row clicks and is reported in section_errors.search_results. Already-read text and independently extracted anchors retain their normal handling. A result without that diagnostic does not guarantee that every conversation was enumerated. |
| send_messageA | Compose and send a new message to a LinkedIn user. Profile-based targeting opens LinkedIn's compose flow. It is not a safe reply path for an existing recruiter/InMail or messaging thread: it may create a separate DM even after you inspect that thread with get_conversation or search_conversations. Those tools only read an existing thread; they do not send a reply. Until a thread-targeted send path is available, do not treat profile-based send_message as a reply. The recipient must be directly messageable from the profile page. If
LinkedIn does not expose a normal Message action, use connect_with_person
first, then retry send_message only after the connection request is
accepted. A
|
| get_feedA | Get posts from the authenticated user's LinkedIn feed. |
| search_postsA | Search LinkedIn posts/content globally by keyword (the "Posts" tab). Use this to catch informal hiring posts ("we're hiring", "Buscamos ...", "estamos contratando", "join our team") that often appear before a formal job listing exists. This is global content search, distinct from get_feed (your own home feed) and get_company_posts (one company's page). |
| comment_on_postA | Comment on a LinkedIn post as the signed-in account. The post is identified by its permalink, and the permalink alone decides
where the comment lands: the tool navigates there and refuses to type
unless the browser's own address is still that post. Pass a reference
exactly as get_feed, search_posts or get_company_posts returned it in
Only the post's own comment box is used, and only while it is the single comment box open on the page and already empty. Text someone else typed is never overwritten, and a reply to an existing comment is not reachable from here. With confirm_comment False this is a read: it reports whether the post
accepts a comment from this account, hands back an Commenting is rate-limited by LinkedIn and automated commenting at volume is what account restrictions are for. Treat each call as a deliberate act the user asked for, not as a loop to run over a list. |
| close_sessionA | Close the current browser session and clean up resources. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 21 tools
Each tool targets a distinct resource/action, and overlapping-seeming areas are well separated by descriptions: get_feed vs search_posts vs get_company_posts, search_people vs get_company_employees, and get_inbox vs get_conversation vs search_conversations. The relationship between send_message and connect_with_person is also clear, despite the reply-path caveat.
Names follow a predictable snake_case verb_noun pattern throughout: get_*, search_*, connect_with_*, send_*, comment_on_*, and close_*. Minor variations like connect_with_person and comment_on_post still fit the convention.
The server covers a broad LinkedIn surface (jobs, profiles, companies, messaging, feed/posts, and session management), so 21 tools is defensible. It is slightly above the ideal 3-15 range but each tool has a clear role.
Coverage is broad across reading and some writing workflows, but there are notable lifecycle gaps: no safe reply to an existing message thread, no job application or save/unsave operation, and no post creation or reaction tools. Agents can still inspect and interact with many resources, but several common LinkedIn workflows are incomplete.