BASH-CMS
Enables governed LinkedIn company-page publishing from a VS Code extension's MCP server, with read-only tools for account info, post preview, and a draft tool for creating pending posts, plus an optional publish tool that is off by default and requires explicit confirmation.
README
BASH-CMS
Governed publishing from your repo, inside VS Code: browse what's worth sharing, draft LinkedIn company-page posts, run the brand guard, approve with your name on it, and publish — with an idempotency ledger so nothing ever double-posts.
Built by BASH Consulting and used to run its own company page. The pipeline follows one rule end to end: drafts are files, approval is a human decision recorded in the file, and a script does the mechanical work. No post leaves your machine without a person deciding it should.
How it works
Content ──▶ Draft (status: pending) ──▶ Approve (status: approved) ──▶ Publish ──▶ Ledger
│ a file in a person, in the POST /rest/posts .github/
.cms/ index drafts/linkedin/ review panel as the org linkedin-log.json
or a frontmatter scan
- Content comes from the
.cms/contract when your repo has one (the zer0-CMS content engine's machine-readable index), and degrades honestly to a frontmatter scan when it doesn't. - Drafts are markdown files with
type/status/sourcefrontmatter — the same queue format the BASH CI lane publishes from, so the extension and a GitHub Actions workflow can share one queue and one ledger without stepping on each other. - The guard is mechanical: LinkedIn's 3,000-character limit, the 140-character "…see more" fold, a banned-phrase list, filler detection, and your own patterns file. Errors block a live post; a human can still be overruled only by a human.
- The ledger is keyed by canonical URL. Either lane (extension or CI) posts first; the other skips. Byte-compatible with the Python lane's
json.dumpoutput, so the file never churns in git.
The four views
| View | What it shows |
|---|---|
| Drafts | The queue, by status — open the review panel, approve, publish |
| Content | Distributable pages (health-ranked when .cms/ exists), with a posted badge |
| What to write next | Catering lanes: A — distribute what exists · B — write more of what landed · C — say the quiet part · D — refresh what worked |
| Published | The ledger, newest first — jump to the live post |
Getting started
- Install the extension, open your content repo.
- Set
bashCms.linkedin.orgUrnandbashCms.site.baseUrl(Settings → BASH-CMS). - Run BASH-CMS: Set LinkedIn Credentials and paste a token with
w_organization_social+r_organization_social(from the token generator; requires Community Management API access on your LinkedIn app). - Run BASH-CMS: LinkedIn: Who Am I to confirm the token, scopes, and expiry.
- Draft from the Content view, review, approve — and when you're ready to go live, enable
bashCms.publish.allow.
Publishing is off by default. Until you enable it, every path — commands, the review panel, MCP — is preview and draft only.
Commands
| Command | What it does |
|---|---|
| Set / Clear LinkedIn Credentials | Token (and optional client id/secret) into VS Code SecretStorage — never settings, never a file |
| LinkedIn: Who Am I | Token status, granted scopes, days to expiry |
| New Article Draft / New Text Draft | Stage a status: pending draft in the queue |
| Preview LinkedIn Payload | The exact /rest/posts body that would send — no network |
| Run Brand Guard on Draft | Errors, warnings, and the fold preview |
| Approve Draft | pending → approved, recorded in the file |
| Publish Draft… / Publish All Approved Drafts | The live path: guard → ledger → confirm → post → record |
| Open Review Panel | The approval gate, made legible (below) |
| Generate Catering Worklist | Write .cms/distribution/worklists/<date>-catering.md |
| Open Post on LinkedIn | Jump from the ledger to the live post |
| Configure MCP Server (.vscode/mcp.json) | Fallback wiring for older VS Code |
The review panel
One panel per draft, opened beside the editor: the commentary with the fold marker where LinkedIn truncates, every guard finding, the exact outgoing JSON, and the ledger state. The Publish button stays disabled until the gates pass — and the extension re-checks every gate on its own side when you click, because a webview is UI, never the gate.
MCP server
The extension registers a bundled MCP server with VS Code (1.101+), so Copilot agent mode — or any MCP client pointed at dist/mcp-server.js — gets six tools:
| Tool | Safe? |
|---|---|
linkedin_whoami · linkedin_list_posts · linkedin_get_post · linkedin_preview |
read-only |
linkedin_draft |
writes a pending draft for a human |
linkedin_publish |
off by default — needs BASH_CMS_MCP_ALLOW_PUBLISH=1 in the server env and confirm: true per call |
The access token is injected into the server process only at start time, from SecretStorage — it never appears in a static definition, a setting, or a log.
Settings
| Setting | Default | Purpose |
|---|---|---|
bashCms.linkedin.orgUrn |
— | The company page (urn:li:organization:…) |
bashCms.linkedin.apiVersion |
202606 |
LinkedIn-Version header (YYYYMM) |
bashCms.site.baseUrl |
— | Canonical site origin |
bashCms.site.permalinkPattern |
/posts/:year/:month/:day/:slug/ |
Canonical URL shape (the ledger key) |
bashCms.content.globs |
pages/_posts/**/*.md |
Where content lives when there's no .cms/ |
bashCms.content.previewsDir / .fallbackImage |
assets/images/… |
Card thumbnail resolution |
bashCms.drafts.folder |
drafts/linkedin |
The queue |
bashCms.ledger.path |
.github/linkedin-log.json |
The shared ledger |
bashCms.publish.allow |
false |
The master gate |
bashCms.publish.acceptStatuses |
pending, approved |
Set to approved only to require the in-editor approval step |
bashCms.guard.bannedPatternsFile |
— | Extra banned patterns ([{name, pattern, flags?}]) |
bashCms.env.readDotEnv |
true |
Read LINKEDIN_* from the workspace .env as a fallback |
Boundaries
Four rules this tool is built around, in order of importance:
- Every post is approved by whoever's name goes on it. There is no unattended path to the live page.
- Each account is authorized by its own owner's consent — a token the owner minted, stored in SecretStorage.
- The audience you write for is declared by you — never derived from connections, followers, or anyone's profile data.
- Everything read back is your own aggregate numbers. No member-level data enters the catering lanes.
Security notes
- Tokens live in VS Code SecretStorage (the OS keychain). The workspace
.envis a read-only fallback for repos that already use one; the extension never writes secrets to disk. - LinkedIn access tokens last 60 days and programmatic refresh is limited to LinkedIn partners — Who Am I tells you the days remaining, and the ledger records the expiry so a CI health-check can warn too.
- The HTTP client never puts the token in an error message or a log line.
Development
npm install
npm run compile # type-check + lint + both bundles
npm test # unit + golden + MCP stdio + integration (downloads VS Code 1.101)
npx @vscode/vsce package # build the .vsix
The test suite includes golden cross-lane fixtures generated by the Python publishing lane (src/test/fixtures/golden/) — the ledger bytes, payload previews, and catering worklists must match that lane byte-for-byte, because both write the same files in the same repos. Regenerate with python3 src/test/fixtures/golden/generate.py <path-to-bashconsultants>.
License
MIT — © Amr Abdel-Motaleb, BASH Consulting.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.