Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
LINKEDIN_MCP_CONFIGNoPath to the configuration JSON file. Overrides config.json resolution. Relative paths are resolved from the working directory.config.json
LINKEDIN_MCP_COUNTERSNoPath to the counters.json file that persists daily action counts. Overrides the default inside the state directory.
LINKEDIN_MCP_LOG_LEVELNoLog level: debug, info, warn, or error. All diagnostics go to stderr.info
LINKEDIN_MCP_READ_ONLYNoSet to '1' to enable read-only mode, equivalent to --read-only. Refuses write tools before any action is attempted. A malformed value (e.g. 'enabled') causes a config_invalid error.0
LINKEDIN_MCP_STATE_DIRNoDirectory for runtime state (storageState.json, chromium-profile, counters, screenshots). Relative paths are resolved from the working directory../.linkedin-mcp
LINKEDIN_MCP_STORAGE_STATENoPath to the Playwright storageState.json file that holds your LinkedIn session. Overrides the default inside the state directory.
LINKEDIN_MCP_USER_DATA_DIRNoPath to the persistent Chromium profile directory. Overrides the default inside the state directory.
LINKEDIN_MCP_SCREENSHOT_DIRNoPath to the directory where diagnostic screenshots are written. Overrides the default inside the state directory.

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

CapabilityDetails
tools
{
  "listChanged": true
}

Tools

Functions exposed to the LLM to take actions

NameDescription
linkedin_loginA

Opens a visible Chromium window on LinkedIn’s login page and waits for the human to sign in themselves, including any 2FA or CAPTCHA step. This tool never types credentials and never reads the password field. When LinkedIn shows a signed-in feed, the browser session is saved to local disk (permissions 0600) and reused by every other tool. Takes no arguments. Not available under --dry-run.

linkedin_session_statusA

Reports whether the locally saved LinkedIn session still works, when it was saved, and when it was last verified. Loads the feed once to check; no cookie or session content is ever returned. Under --dry-run this always reports valid. Takes no arguments.

linkedin_create_postA

Publishes a text post to the signed-in member’s LinkedIn feed. This is a two-step tool: called without confirm: true it only returns a preview (character count, line count, audience, and a previewToken) and touches nothing on LinkedIn; call it again with the same arguments plus confirm: true to actually publish. Counts against the daily posts cap. visibility defaults to "public" (Anyone); "connections" restricts the post to 1st-degree connections. LIMITATION: mediaUrl is accepted for validation but images and video are NOT supported — this server never performs its own network fetches, so it cannot download remote media, and confirming a call that sets mediaUrl is refused with invalid_input rather than silently publishing text only. Post without media, or attach the image by hand in LinkedIn afterwards. Posts over roughly 1300 characters are collapsed by LinkedIn behind "…see more" (a warning, not an error). Under --dry-run the composer is driven against a local HTML fixture and the final "Post" click is skipped.

linkedin_send_connection_requestA

Sends one connection invitation to a member, optionally with a note. Two-call handshake: without confirm: true this only reads the profile (name, headline, connection degree, and whether a Connect control exists at all) and returns a preview plus a previewToken — nothing is clicked and no invitation is sent. Re-issue the same call with confirm: true to send. profileUrl accepts a slug ("john-doe") or any LinkedIn profile URL. note is capped at 300 characters; free accounts are limited to 200 and LinkedIn silently truncates past its own limit, so notes over 200 characters are flagged. The request is REFUSED (with invalid_input, before any quota is spent) when the member is already a 1st-degree connection, when an invitation is already pending, or when LinkedIn offers no Connect control for that profile; the preview says so up front in that case. Counts against the daily connectionRequests cap. Under --dry-run the invitation dialog is driven against a local HTML fixture and the final Send click is skipped.

linkedin_send_messageA

Sends one direct message, either to a 1st-degree connection (pass a profile URL or slug) or into a conversation that already exists (pass the id from its /messaging/thread// URL). Two-call handshake: without confirm: true this only reads the profile or thread and returns a preview plus a previewToken, and nothing is typed or sent. Messaging is limited to 1st-degree connections — a profile at 2nd/3rd degree (or one whose degree cannot be read) is refused with not_connected, and InMail is never attempted. Counts against the daily messages cap. Under --dry-run everything runs against local fixtures and the final Send click is skipped.

linkedin_scrape_profileA

Reads a LinkedIn member profile and returns it as structured data: name, headline, about, location, connection degree, experience, education and skills. profileUrl accepts a full URL ("https://www.linkedin.com/in/john-doe"), a path ("/in/john-doe") or a bare slug ("john-doe"); omit it entirely to read the signed-in member's own profile. Read-only: takes no confirm, consumes no daily quota, and the only thing it clicks is the "…see more" expander, which reveals text already on the page. Profile sections LinkedIn hides from the viewer come back as null or [] rather than as an error — only a profile with no readable name fails (selector_not_found), because that means the page is not a profile. connectionDegree is 1, 2 or 3, and null when no badge is shown (which includes your own profile). skills is capped at 50 entries, experience at 25, education at 15.

linkedin_scrape_feedA

Reads the signed-in member's home feed and returns the posts as structured data: author name and headline, post text, permalink, reaction and comment counts, and a timestamp. count is how many posts to return (default 10, max 50). The feed is lazy-loaded, so reaching a high count means scrolling, and this tool pauses a randomized human-like interval between scrolls — a large count can therefore take a minute or more. Read-only: takes no confirm, consumes no daily quota, and never likes, comments or reposts; the only clicks are "…see more" expanders so post text is captured at full length. Posts are de-duplicated by permalink (LinkedIn recycles cards while scrolling) and sponsored or suggestion cards with no readable author are skipped. Fewer posts than requested is a normal result, not an error.

linkedin_search_jobsA

Searches LinkedIn job postings and returns them as structured data: job id, title, company, location, whether the posting supports Easy Apply, and a URL. keywords is required; location is a free-text place name ("London", "Remote", "New York, NY"). Four optional facets narrow the search — easyApplyOnly, datePosted ("past24h" | "pastWeek" | "pastMonth"), experienceLevel ("internship" | "entry" | "associate" | "midSenior" | "director" | "executive") and remote — each accepted either at the top level or nested inside a filters object; filters wins if you somehow pass both. count is how many postings to return (default 25, max 100); results are lazy-loaded, so a high count means scrolling with a randomized human-like pause between scrolls and can take a while. Read-only: takes no confirm and consumes no daily quota. Cards LinkedIn renders without a usable job id are skipped and counted in skipped. Pass a returned jobId to linkedin_apply_to_job, which handles Easy Apply postings only — so filter on easyApply before trying to apply.

linkedin_apply_to_jobA

Submits a LinkedIn Easy Apply application. jobId accepts a bare id ("3812345678"), a job URL, a ?currentJobId= URL or a urn:li:jobPosting: URN. resumePath is optional — an absolute path to a local .pdf, .doc or .docx file (never uploaded anywhere except LinkedIn itself); omit it to use whatever resume is already attached to your LinkedIn profile. answers maps question text to the answer to give; keys are matched case-insensitively and by substring, so "years of experience" answers "How many years of experience do you have?". CONFIRMATION IS REQUIRED: called without confirm: true this opens the Easy Apply modal read-only, lists every question it will be asked, flags the required ones it cannot answer in unansweredRequired, closes the modal without submitting, and returns a previewToken. Re-issue the identical call with confirm: true to actually submit. Postings that apply through an employer's own site are refused with external_application and their URL handed back for you to use by hand; postings with no Easy Apply button are refused with not_easy_apply. If a required question still cannot be answered at submit time the application is abandoned with invalid_input rather than sent incomplete — this tool never submits a partial application. Counts against the daily jobApplications cap.

linkedin_list_pending_invitesA

Lists invitations that are still awaiting a decision. direction: "received" (the default) returns invitations other people sent you; direction: "sent" returns invitations you sent that have not been accepted yet. count caps how many are returned (default 25, max 100). Read-only: this never accepts, ignores or withdraws anything, and takes no confirm. Each entry carries name, headline, canonical profile URL, and a timestamp (the datetime attribute when LinkedIn provides one, otherwise the relative text such as "3 days ago"). An empty list is a normal result, not an error. total is how many invitation cards were found on the page, which can exceed returned when count is smaller.

linkedin_list_connectionsA

Lists the signed-in member’s 1st-degree connections, most recently added first (LinkedIn’s default sort). count caps how many are returned (default 50, max 200) and the list is scroll-paginated until that many are loaded, the list stops growing, or an internal scroll cap is reached. query is an optional case-insensitive substring filter applied LOCALLY to the name and headline of the connections that were loaded — it is not LinkedIn’s own search, so a match that sits far down a very long list may not be reached. Read-only: takes no confirm and changes nothing. Each entry carries name, headline, canonical profile URL, and a "connected at" timestamp when LinkedIn exposes one. total is the account-wide count from the page header when it can be read (totalSource: "page-header"), otherwise the number of cards that were loaded.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/bansalsahab/linkdin-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server