All projects

Developer Tools · MCP

LinkedIn Profile MCP

An offline MCP server I published on npm that reviews LinkedIn headline, About and Experience text the way technical recruiters search, with no login or scraping.

Context
Open-source MCP server · my solo project, published on npm
Stack
TypeScript, MCP, Vitest

How it works

Recorded terminal session, replayed

Recorded MCP session, replayed: real linkedin-profile-mcp 0.1.1 tool output on a fictional sample profile.

Solo project

  1. List the tools

    The client connects over stdio and calls tools/list. The server returns its 6 tools: get_guidelines, lint_field, review_headline, review_experience, keyword_coverage, review_profile.

  2. Lint a headline

    lint_field on a fictional headline flags an em dash and several stock phrases, and shows what is visible in the first 60 characters.

  3. Review the headline

    review_headline adds pass/fail structure checks: role in the first 60 characters, character use, segments, and an open-to line.

  4. Review an Experience entry

    review_experience checks a short bullet list: bullet count, how many bullets carry a number, and whether each starts with a verb.

  5. Keyword coverage

    keyword_coverage compares the profile with two fictional job postings and reports which recurring terms are covered and which are missing (for example dbt). The two prompts are then listed.

Problem

Recruiters find people with keyword searches, and a search result shows only the opening of a headline, so the wording of a profile decides what a search can match. Guidance on this is scattered, and tools that act on a profile usually want a login or scrape the page. After a senior technical recruiter's talk I wanted that guidance as checks an AI assistant can run on text I paste in.

My contribution

This is my solo project, and the repository's history has one author. I wrote the rules engine, the six tools, the two prompts and two resources, the tests, and the packaging for npm and the official MCP Registry. The recruiter guidance behind the rules comes from that talk.

Approach

linkedin-profile-mcp is a TypeScript server built on the Model Context Protocol SDK, with zod schemas for its inputs. It never logs in, scrapes, posts or calls the network: you paste text in and copy results out, and a pure, offline, deterministic rules engine does the checking. lint_field flags stock phrases and dashes and shows what is visible in the first 60 characters. review_headline and review_experience add structure checks, such as bullet count, how many bullets carry a number and whether each starts with a verb. keyword_coverage compares a profile with job postings. get_guidelines and review_profile complete the six tools, and the optimize_section and full_profile_review prompts and the linkedin://guide and linkedin://limits resources round out the server.

How it was built

I built linkedin-profile-mcp in TypeScript on the Model Context Protocol SDK, with zod validating each tool's input. The checks live in one pure rules module, so the same text always gives the same result, and a Vitest file of 9 tests covers it. The server exposes six tools, two prompts and two resources over stdio. I kept it offline on purpose: it has no login, scraping, posting or network calls. I published it to npm, the official MCP Registry and Glama. Limits: it checks wording and structure, not whether a profile is true, and I have no usage data.

Evidence and limitations

Vitest runs 9 tests in one file and all pass. The package is version 0.1.1 under the MIT licence, published to npm on 8 Nov 2026, and listed in the official MCP Registry as io.github.Ruthvik-Bandari/linkedin-profile-mcp and on Glama.

The recording above replays a real session against a fictional sample profile and shows four of the six tools. review_profile, get_guidelines and the prompts' text are not shown.

Limits: the checks are rules, so they judge wording, structure and keyword coverage and cannot tell whether a profile is accurate or will draw replies. I have no usage or outcome data and make no claim about adoption.