Source of truth for Verana (and 2060.io) social media content and the veranafoundation.org blog, organized one folder per brand:
social-posts/
├── 2060.io/ # 2060.io brand — posting rules
├── verana/ # Verana social posts (LinkedIn / X) + assets
└── veranafoundation.org/ # Blog posts rendered live by the Foundation website
Public repo. This repository is public (the Foundation website fetches the blog content from it). Never commit secrets — API keys live only in
~/.mcp/*.env, outside this repo.
| Path | What it is |
|---|---|
post-rules.md |
Repo-wide rules: file naming, required sections, writing style. |
verana/verana-post-rules.md |
Verana voice, sources of truth, terminology guardrails. |
2060.io/2060.io-post-rules.md |
2060.io posting rules. |
verana/YYYYMMDD-NNN.md (+ images) |
Individual social posts with their generated media. |
verana/org-page-copy.md |
LinkedIn org-page copy (tagline, overview). |
verana/assets/ |
Brand assets (e.g. the LinkedIn banner + its HTML source). |
veranafoundation.org/blog/ |
Live blog content the website renders (see its README). |
scripts/publish.mjs · scripts/PUBLISHING.md |
Publishing script + playbook (see Publishing). |
Posts are published through two MCP (Model Context Protocol) servers driven by Claude Code or Claude Desktop:
| Server | Image | What it does |
|---|---|---|
| linkedin-mcp | io2060/linkedin-mcp:latest |
Publish + engage on LinkedIn as the authenticated member (posts, image posts, reshares, comments, reactions). |
| x-autonomous-mcp | io2060/x-autonomous-mcp:latest |
Full X (Twitter) access — post, search, read timelines, like, retweet, upload media. |
Both servers run as stdio MCP servers inside Docker, with credentials supplied via
an --env-file and persistent state mounted on a host volume. This README explains how
to set that up locally.
Source repos:
2060-io/linkedin-mcp·2060-io/x-autonomous-mcp
- Docker installed and running (the MCP servers run as containers — no Node toolchain needed locally).
- Claude Code (CLI) and/or Claude Desktop installed.
- A LinkedIn developer app (for linkedin-mcp) and X developer app (for x-autonomous-mcp). See Getting credentials below.
Pull the images once:
docker pull io2060/linkedin-mcp:latest
docker pull io2060/x-autonomous-mcp:latestBoth servers read their secrets from env files and persist tokens/budgets/dedup state to
a mounted data directory. Create them under ~/.mcp:
mkdir -p ~/.mcp/linkedin-data ~/.mcp/x-data# --- LinkedIn OAuth app (from the LinkedIn Developer Portal) ---
LINKEDIN_CLIENT_ID=your_client_id
LINKEDIN_CLIENT_SECRET=your_client_secret
LINKEDIN_REDIRECT_URI=http://localhost:8000/oauth/callback
LINKEDIN_API_VERSION=202506
# --- Basic-Auth protecting the OAuth callback routes ---
LI_MCP_ADMIN_USER=admin
LI_MCP_ADMIN_PASSWORD=choose_a_strong_password
# State file path INSIDE the container (maps to ~/.mcp/linkedin-data on the host)
LI_MCP_STATE_FILE=/data/linkedin-mcp-state.json
# --- Optional daily budget limits (defaults shown) ---
# LI_MCP_MAX_POSTS=5
# LI_MCP_MAX_COMMENTS=10
# LI_MCP_MAX_REACTIONS=30
# LI_MCP_MAX_DELETES=3LINKEDIN_REDIRECT_URI must exactly match a Redirect URL configured on your LinkedIn
app. Set a budget value to 0 to disable that action entirely, or -1 for unlimited.
# --- X (Twitter) API credentials (5 values from the X Developer Portal) ---
X_API_KEY=your_consumer_key
X_API_SECRET=your_secret_key
X_BEARER_TOKEN=your_bearer_token
X_ACCESS_TOKEN=your_access_token
X_ACCESS_TOKEN_SECRET=your_access_token_secret
# --- Optional daily budget limits (defaults shown) ---
# X_MCP_MAX_REPLIES=8
# X_MCP_MAX_ORIGINALS=2
# X_MCP_MAX_LIKES=20
# X_MCP_MAX_RETWEETS=5
# X_MCP_MAX_FOLLOWS=10
# X_MCP_MAX_UNFOLLOWS=10
# X_MCP_MAX_DELETES=5Keep these files private. They contain live API secrets — never commit them.
~/.mcplives outside this repo on purpose.
- Go to the LinkedIn Developer Portal and create an app — it must be linked to a Company Page.
- On the Products tab, request Sign In with LinkedIn using OpenID Connect and
Share on LinkedIn (grants
openid,profile,email,w_member_social). - On the Auth tab, copy the Client ID / Client Secret, and add
http://localhost:8000/oauth/callbackas a Redirect URL (must matchLINKEDIN_REDIRECT_URI).
- Go to the X Developer Portal → Apps → Create App.
- Under User authentication settings → Set up: set App permissions to
Read and write, Type of App to Web App, Automated App or Bot, Callback
URI to
https://localhost, and any Website URL. - On Keys and Tokens, copy API Key, API Secret, and Bearer Token, then Regenerate the Access Token and Access Token Secret after enabling write permissions (otherwise tokens are read-only and posting fails with 403).
X uses long-lived API tokens (already in x.env — nothing else to do). LinkedIn needs a
one-time OAuth login to obtain member tokens, which are then persisted to
~/.mcp/linkedin-data/linkedin-mcp-state.json.
Run the server in HTTP mode just for the login, complete the consent flow, then stop it:
docker run --rm -i \
--env-file ~/.mcp/linkedin.env \
-v ~/.mcp/linkedin-data:/data \
-e MCP_TRANSPORT=http -e MCP_PORT=8000 \
-p 8000:8000 \
io2060/linkedin-mcp:latestThen in a browser open http://localhost:8000/oauth/start, authenticate with the Basic-Auth
credentials from linkedin.env (LI_MCP_ADMIN_USER / LI_MCP_ADMIN_PASSWORD), approve on
LinkedIn, and you'll be redirected back. Check http://localhost:8000/oauth/status to confirm
a valid token is stored, then stop the container (Ctrl-C). The tokens now live on the
mounted volume and are reused by the stdio server below.
The easiest way is claude mcp add (the -- separates the claude flags from the command
Claude runs). Use --scope user to make the servers available in every project:
# LinkedIn
claude mcp add linkedin --scope user -- \
docker run --rm -i \
--env-file "$HOME/.mcp/linkedin.env" \
-v "$HOME/.mcp/linkedin-data:/data" \
-e MCP_TRANSPORT=stdio \
io2060/linkedin-mcp:latest
# X (Twitter)
claude mcp add x-twitter --scope user -- \
docker run --rm -i \
--env-file "$HOME/.mcp/x.env" \
-v "$HOME/.mcp/x-data:/data" \
-e MCP_TRANSPORT=stdio \
io2060/x-autonomous-mcp:latestVerify and restart Claude Code afterwards:
claude mcp list # both should show ✔ ConnectedEquivalent manual config (~/.claude.json under mcpServers)
{
"mcpServers": {
"linkedin": {
"type": "stdio",
"command": "/usr/local/bin/docker",
"args": [
"run", "--rm", "-i",
"--env-file", "/Users/<you>/.mcp/linkedin.env",
"-v", "/Users/<you>/.mcp/linkedin-data:/data",
"-e", "MCP_TRANSPORT=stdio",
"io2060/linkedin-mcp:latest"
],
"env": {}
},
"x-twitter": {
"type": "stdio",
"command": "/usr/local/bin/docker",
"args": [
"run", "--rm", "-i",
"--env-file", "/Users/<you>/.mcp/x.env",
"-v", "/Users/<you>/.mcp/x-data:/data",
"-e", "MCP_TRANSPORT=stdio",
"io2060/x-autonomous-mcp:latest"
],
"env": {}
}
}
}Use absolute paths here — ~ and $HOME are not expanded inside args. Find your
Docker path with which docker.
Claude Desktop reads MCP servers from its config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
Edit it (create it if missing) and add the same two servers. Claude Desktop does not
inherit your shell PATH, so use the absolute Docker path and absolute env-file /
volume paths:
{
"mcpServers": {
"linkedin": {
"command": "/usr/local/bin/docker",
"args": [
"run", "--rm", "-i",
"--env-file", "/Users/<you>/.mcp/linkedin.env",
"-v", "/Users/<you>/.mcp/linkedin-data:/data",
"-e", "MCP_TRANSPORT=stdio",
"io2060/linkedin-mcp:latest"
]
},
"x-twitter": {
"command": "/usr/local/bin/docker",
"args": [
"run", "--rm", "-i",
"--env-file", "/Users/<you>/.mcp/x.env",
"-v", "/Users/<you>/.mcp/x-data:/data",
"-e", "MCP_TRANSPORT=stdio",
"io2060/x-autonomous-mcp:latest"
]
}
}
}Replace /Users/<you> with your home directory (/Users/mj here), and /usr/local/bin/docker
with the output of which docker if different. Fully quit and reopen Claude Desktop —
the servers appear under the 🔌 (MCP) icon when connected.
In either client, ask:
- LinkedIn: "Who am I posting as on LinkedIn?" → runs
get_me, returns your profile and remaining daily budget. - X: "What's my X budget right now?" → returns the budget counters.
Both servers report a daily budget in every response (e.g.
li_budget: "0/5 posts used, ..." / x_budget: "0/8 replies used, ..."), so the assistant
always sees its remaining limits.
Both servers enforce hard per-day limits server-side — they refuse over-budget actions
even if the model is told to ignore them — plus engagement deduplication (never like /
comment / retweet the same post twice). Tune limits via the *_MAX_* env vars in the
respective env file (0 = disabled, -1 = unlimited). See each source repo's README for
the full option list.
Posts are published with scripts/publish.mjs, which drives the
MCP servers over stdio JSON-RPC and reads any image from disk (so large media never has
to pass through a chat tool-call argument). The full procedure — and the hard-won gotchas
(X's 280-char limit returning a misleading 403, & → & escaping, image compression
with sips, stripping Markdown) — is documented in
scripts/PUBLISHING.md.
# LinkedIn (image, public) + X (image) — separate text per platform
node scripts/publish.mjs --linkedin --image verana/PIC.jpg --visibility PUBLIC \
--alt "…" --text-file verana/.li-text.txt
node scripts/publish.mjs --x --image verana/PIC.png --text-file verana/.x-text-280.txtAlways write posts per post-rules.md (and
verana/verana-post-rules.md for Verana): at most one em
dash, no Markdown / & in the body, X variants ≤ 280 chars.
veranafoundation.org/blog/ is the source of truth for
the Blog on veranafoundation.org/blog. The website
fetches these Markdown posts from this repo over the GitHub API and renders them with ISR —
this repo is the content store; the site is just the renderer. Each post is a Markdown file
with YAML front matter (title, date, tag, excerpt, author, authorAvatar,
authorSocial, …). See the blog README for the
authoring format.
| Symptom | Fix |
|---|---|
claude mcp list shows the server but tools aren't available |
Restart the client — tool lists load at startup. |
| Server won't connect | Confirm Docker is running and the image is pulled; run the docker run … command by hand to see the error. |
Claude Desktop can't find docker |
Use the absolute path from which docker; Desktop doesn't inherit your shell PATH. |
| LinkedIn calls fail with auth errors | Re-run the one-time OAuth; check /oauth/status. |
| X posting returns 403 oauth1-permissions | Access Token was created before write perms were enabled — set the app to Read and write and Regenerate the token. |
| X returns 401 Unauthorized | First suspect a stale container (see next row) — especially right after regenerating keys. Otherwise re-check all 5 credentials for typos / stray spaces, and confirm the Consumer Key/Secret and Access Token/Secret are from the same app. |
Edited an *.env but the server still uses the old credentials (e.g. 401 after regenerating keys) |
These stdio servers read the env file at container start. A long-lived MCP connection keeps a container running, so it holds the old values. Reconnect the server (/mcp → reconnect) or kill the running container so the next call re-reads the env: docker kill $(docker ps -q --filter ancestor=io2060/x-autonomous-mcp:latest). |
| Action refused with a budget message | The daily limit is exhausted — raise the *_MAX_* value or wait for the daily reset. |
social-posts/
├── post-rules.md # repo-wide posting rules
├── 2060.io/
│ └── 2060.io-post-rules.md
├── verana/
│ ├── verana-post-rules.md # Verana voice + sources of truth
│ ├── org-page-copy.md # LinkedIn org-page tagline + overview
│ ├── YYYYMMDD-NNN.md (+ images) # individual social posts
│ └── assets/ # banner.html + rendered banner PNG
├── veranafoundation.org/
│ └── blog/ # live blog content (+ its own README)
├── scripts/
│ ├── publish.mjs # publish to LinkedIn / X via the MCP servers
│ └── PUBLISHING.md # publishing playbook
├── LICENSE
└── README.md