Draft social posts from a chat app, with each brand in its own container

Draft social posts from WhatsApp or Telegram with NanoClaw

NanoClaw is a free, open-source assistant you run on your own machine and message from chat apps such as WhatsApp, Telegram, Slack and Discord. Add Bolta as a remote MCP server on a NanoClaw agent group and it can draft posts in your brand voice and leave them for your yes.

  • Message your own NanoClaw agent in a chat app and get a draft back in your brand voice
  • Add Bolta to one agent group at a time, so each brand or client stays in its own container
  • Your Bolta key belongs in NanoClaw's vault, and drafts wait for your yes

Nothing is approved, scheduled or published until you say yes.

Partly confirmed, see what is and is notChecked 2026-11-21

NanoClaw runs on your own hardware, so you add the server address https://mcp.bolta.ai/mcp on that machine.

NanoClawBolta connected
You

Draft three LinkedIn posts about our launch and send them to review

Illustration of the reply: Bolta returns a draft showing the connected account and the scheduled time, with Approve, Edit and Reject actions.

Illustration. Your drafts show your real text, account and time.
Quick connect

One command on the host, then ask from your phone

  1. 01

    Add Bolta to a group and restart it

    On the machine running NanoClaw, run ncl groups config add-mcp-server --id <group-id> --name bolta --url https://mcp.bolta.ai/mcp, then ncl groups restart --id <group-id>. Put your Bolta key in the credential vault, not in the command.

  2. 02

    Ask in your chat app

    Message the agent: list my Bolta workspaces and accounts. Then ask for drafts. They land in review and wait for your yes.

What is and is not confirmed

Read this before you set up NanoClaw

  • Confirmed from NanoClaw's official docs: you can add a remote Streamable HTTP MCP server to an agent group with ncl groups config add-mcp-server --id <group-id> --name <name> --url <https url>. The URL must be HTTPS and cannot carry credentials, which https://mcp.bolta.ai/mcp satisfies.
  • Confirmed from the docs: --headers is for non-secret values only, and real credentials belong in the credential gateway vault (OneCLI Agent Vault by default), which injects them in flight. The docs also say not to put a raw key in --env.
  • Not confirmed: the exact vault command for storing a Bolta key, and a worked example of the vault injecting a Bearer bolta_sk_ key for a custom MCP host. This page does not write out a one-line Authorization header, because none is documented.
  • Not confirmed: whether NanoClaw's remote MCP servers support OAuth sign-in, and whether Bolta's skill works inside its agent containers. Neither has been run end to end.
  • Whatever works on the NanoClaw side, Bolta does not change: nothing is approved, scheduled or published until you say yes.
The flow

What happens when you message your NanoClaw agent about Bolta

Each step is a Bolta tool the agent calls through the MCP server you added to its group. Nothing here approves, schedules or publishes anything.

  1. 01

    Your message reaches the agent group

    You write from WhatsApp, Telegram, Slack or Discord. NanoClaw starts or reuses the containers for the group you are talking to, and that group is the only one with Bolta attached.

  2. 02

    It checks which brand you mean

    The agent lists your workspaces so a draft for one brand is not written under another. If you have several, it asks before going further.

    list-workspaces

  3. 03

    It finds the accounts it could draft for

    It reads the networks connected to that workspace. If none are connected, it asks Bolta for a connect link instead of making one up.

    list-accounts

  4. 04

    It reads your voice and the posting rules

    Before writing, it pulls the saved voice context and the rules for the network, so length and tone match what you have already set up.

    get-voice-context

  5. 05

    It saves the draft, nothing more

    The text is stored in Bolta as a draft. It is not approved, not scheduled and not published.

    save-post

  6. 06

    You decide in chat or in Bolta

    The agent shows the exact text in the chat. Sending it to review, approving, scheduling or publishing happens only after you say yes for that post.

Setup

Set up NanoClaw

Connect once. Everything below works from NanoClaw afterwards.

  1. 01

    Install NanoClaw if you have not

    Clone github.com/nanocoai/nanoclaw and run bash nanoclaw.sh. You need Docker, pnpm 10 or newer and a current Node.js, so check the README for exact versions. Each agent group runs in its own containers.

  2. 02

    Pair a chat app

    Use NanoClaw's /add-<channel> setup for WhatsApp, Telegram, Slack or Discord so you can message your agent. Do this before adding Bolta, so you can test the result from your phone.

  3. 03

    Connect your social accounts in Bolta

    Open Bolta and connect the networks you post to. Without them the agent has nothing to post to, and it will tell you so instead of guessing.

  4. 04

    Create a Bolta API key

    In Bolta, open Settings, then API Keys, and create a key that starts with bolta_sk_. Pick the brand it is for and give it only the permissions you need.

  5. 05

    Put the key in NanoClaw's vault

    Store it in the credential gateway vault (OneCLI Agent Vault by default) so the key is injected when the agent calls Bolta. Do not paste it into --env or --headers: NanoClaw's docs say --headers is for non-secret values and not to put a raw key in --env. The exact vault command is not confirmed here, so follow NanoClaw's own vault docs.

  6. 06

    Add Bolta to your agent group

    From the host shell, find your group id and run ncl groups config add-mcp-server --id <group-id> --name bolta --url https://mcp.bolta.ai/mcp. Run it on the host, not from inside an agent container, where write commands are approval-gated.

  7. 07

    Restart the group

    Changes apply on the next container spawn. Run ncl groups restart --id <group-id> so the new server loads now.

  8. 08

    Check what it can see

    Message your agent: list my Bolta workspaces and accounts. Confirm it shows the right brands and only the ones you want it to use.

  9. 09

    Ask for a draft

    Try: draft three LinkedIn posts about our launch and send them to review. Drafts are free. Approve, schedule or publish them in Bolta, or say yes in chat after the agent shows the exact text.

Connect

Connect NanoClaw to Bolta

NanoClaw runs on your own hardware, so you add the server address https://mcp.bolta.ai/mcp on that machine.

Connection methods

Which Bolta connection methods work with NanoClaw

Which Bolta connection methods work with NanoClaw
MethodWith NanoClawHow
Remote MCP server added with nclNot confirmedNanoClaw's docs describe ncl groups config add-mcp-server with an HTTPS URL, so https://mcp.bolta.ai/mcp can be registered on a group, then restart the group. The sign-in step is not documented for this host, so the full route has not been run end to end.
MCP with an API key in the credential vaultNot confirmedThe docs say credentials belong in the vault, not in --env or --headers. The vault command for a bolta_sk_ key, and the injection for a custom MCP host, are not documented, so this has not been run end to end.
Hosted MCP with OAuth sign-inNot confirmedNanoClaw's docs do not describe OAuth for remote MCP servers, so this page does not promise a sign-in screen.
Skill via npx skills addNot confirmedNanoClaw documents skills as SKILL.md workflows and read-only container skills. An npx skills install path is not described, and Bolta's skill has not been tried in its containers.
Plugin or extensionNot availableBolta does not publish a plugin for NanoClaw. The MCP server is the documented way in.
Bolta REST APINot confirmedBolta has a public API, but nobody has wired it into a NanoClaw group and checked it. The MCP route is the one this page follows.
Capabilities

What you can run in Bolta from NanoClaw

Each row is an area of Bolta, with the real tools behind it.

  • Drafts from a phone message

    Ask for a post in one line and the agent saves it to Bolta as a draft in your brand voice, then shows you the text.

    get-voice-contextget-posting-rulessave-post

  • Review queue from chat

    Ask what is waiting and the agent reads your inbox. Sending drafts to review or approving them happens only after you say yes for that task.

    list-inbox-itemssubmit-for-reviewapprove-many-posts

  • Weekly drafting as a routine

    A repeating job is planned first, saved switched off, and only turned on after a separate yes from you.

    planconfirm-plancreate-routine

  • One brand per agent group

    Each group runs in its own containers with its own memory, and you can name the workspace in every request so drafts do not cross brands.

    list-workspaceslist-accounts

  • Results before you plan

    Ask how last month went and the agent reads the figures Bolta holds for the connected accounts. It does not invent numbers when there are none.

    get-account-analyticslist-workspace-posts

  • Connecting a missing network

    If a network is not linked yet, the agent asks Bolta for a connect link you open yourself, rather than guessing at an account.

    get-connect-link

In practice

A sample chat with a NanoClaw agent on Telegram

An illustration of the order things happen in. It is not real output.

IllustrationNot real output
  1. You

    Draft a LinkedIn post about our new onboarding checklist for the Acme brand.

  2. Tool call
    list-workspaces

    Finds the Acme workspace

  3. Tool call
    list-accounts

    Checks a LinkedIn account is connected

  4. Tool call
    get-voice-context

    Reads the saved voice for Acme

  5. Tool call
    get-posting-rules

    Reads the LinkedIn rules

  6. Tool call
    save-post

    Saves one draft. Not approved, not scheduled

  7. NanoClaw

    I saved a draft for Acme on LinkedIn: a short post about the checklist with one question at the end. Nothing is approved or published. Want the exact text here?

  8. You

    Yes, show me.

  9. NanoClaw

    Here is the text: [post text shown in full]. It is still a draft. Tell me if you want changes, or approve it in Bolta when you are ready.

  10. You

    Make the opening shorter. I will approve it from my laptop later.

  11. Tool call
    save-post

    Edits the same draft. Still not approved or scheduled

  12. NanoClaw

    Done, the draft has a shorter opening and is still waiting. I have not approved or published it.

Which route

Hosted MCP server or a custom setup

NanoClaw can reach Bolta in one way that is documented on both sides.

Hosted MCP server

Choose this for almost everyone.

Point a group at https://mcp.bolta.ai/mcp with ncl groups config add-mcp-server. Bolta keeps its own tools and approval rules, and you update nothing when they change.

Bolta skill in a coding harness

Optional, and unconfirmed inside NanoClaw.

NanoClaw documents skills as SKILL.md workflows for the host and container skills that mount read-only. An npx skills install path is not documented there, so treat the Bolta skill as an extra, not the way in.

Use cases

What can NanoClaw do with Bolta?

  • Send a one-line message from your phone and get a LinkedIn draft back before you leave the train
  • Give each client brand its own agent group, so each client stays in its own containers
  • Have a scheduled chat job ask for next week's drafts every Friday, with approval still in Bolta
  • Ask in Slack or Discord what is waiting for approval, then open the list in Bolta
  • Run the agent on your own hardware, with the Bolta key held in the vault instead of in the agent
  • Ask for last month's post results in chat before you plan the next batch
The connection

What is the Bolta MCP?

The server NanoClaw talks to when it works with Bolta.

MCP, the Model Context Protocol, is the open standard that lets an AI app call tools in another product. Bolta runs a hosted MCP server at https://mcp.bolta.ai/mcp. Your AI app discovers its 67 tools and calls them for you: list your brands, save a draft, plan a routine, read results.

One connection is one organization. You tick the brands the app may reach, and every call is checked against your role, so the app can never do more in Bolta than you can. Approving, scheduling and publishing need your yes.

  • Brands & accounts6 tools

    The brands this connection reaches, who works on them and who can approve, their approval settings, connected social accounts, and each platform's posting rules.

    list-workspaces, list-workspace-members, set-brand-approval, list-accounts, get-connect-link, get-posting-rules

  • Posts9 tools

    Save a draft or edit a post with save-post, then publish it now or at a time with publish-post after the user says yes.

    list-workspace-posts, get-post, save-post, publish-post, delete-post, bulk-create-posts, create-wordpress-blog-post, list-recently-deleted, restore-deleted

  • Review4 tools

    Everything waiting for a yes: submit drafts to the team or send them to a named person for approval, approve one or many, reject with a reason, and work the inbox of routine and campaign drafts.

    list-inbox-items, submit-for-review, approve-many-posts, reject-post

  • Plans6 tools

    Have Bolta plan a routine or a campaign the way its own chat does, change it, then save it on the user's yes.

    plan, plan-routine, plan-campaign, recommend-routines, revise-plan, confirm-plan

  • Routines10 tools

    Work that repeats and runs in Bolta even with the app closed.

    list-routine-templates, list-routines, get-routine, create-routine, update-routine, activate-routine, pause-routine, archive-routine, run-routine-now, list-routine-runs

  • Campaigns6 tools

    One to four weeks of posts about one thing: a launch, an event, a season.

    list-content-campaigns, get-content-campaign, create-content-campaign, update-content-campaign, analyze-content-campaign-source, set-content-campaign-status

  • Team versions4 tools

    One idea written as one post per person on the team, each in their own voice.

    plan-team-posts, update-team-post-plan, generate-team-post-drafts, get-team-post-plan

  • Voice & business8 tools

    The brand's voice and business facts, and writing in that voice.

    get-voice-context, get-voice-profile, create-voice-profile, update-voice-profile, list-business-dna, extract-business-dna, update-business-dna, voice-generate

  • Suggestions2 tools

    What Bolta suggests changing, for a routine or the whole brand, from what people approved, edited and rejected.

    list-suggestions, decide-suggestion

  • Replies10 tools

    Find conversations worth joining, draft replies in the brand voice, and send them once the user says yes.

    search-engagement-opportunities, list-engagements, get-social-content, draft-engagement-reply, draft-inbound-reply, publish-engagement-reply, reply-to-social-content, react-to-social-content, send-outreach-reply, update-outreach-reply

  • Results2 tools

    How posts and accounts are performing.

    list-account-insights, get-account-analytics

Try it from your chat app

Seven messages to send your NanoClaw agent

Illustrative replies. Real output depends on your account.

You in NanoClaw

Draft a LinkedIn post about our spring release.

Bolta

I saved one LinkedIn draft in your brand voice. Nothing is approved, scheduled or published. Want me to show the exact text here before you decide?

Draft saved. Waits for your yes.

save-post

You in NanoClaw

What is waiting for me in Bolta?

Bolta

I read your Bolta inbox. It lists drafts waiting for review and any replies Bolta drafted for you. I have not approved or sent anything.

Read only. Nothing changed.

list-inbox-items

You in NanoClaw

Which brands and accounts can you reach?

Bolta

I can see your workspaces and the accounts connected to each. Tell me which brand to use, or I will ask before drafting.

Read only. Nothing changed.

list-workspaces, list-accounts

You in NanoClaw

Every Friday morning, draft next week's posts for this brand.

Bolta

Here is the plan: weekly drafts, sent to review, not auto-posted. Should I save it switched off? I will only turn it on after a separate yes from you.

Plan shown. Waits for your yes.

plan, confirm-plan, create-routine

You in NanoClaw

Show me last month's results for the Acme brand before I plan the next batch.

Bolta

I can read the results Bolta holds for the connected accounts and summarise them here. If an account has no data yet I will say so rather than guess.

Read only. Nothing changed.

get-account-analytics, list-workspace-posts

You in NanoClaw

Do I have a Bluesky account connected for this brand?

Bolta

Here is what is connected to that workspace. If Bluesky is missing, I can ask Bolta for a connect link that you open yourself.

Read only. Nothing changed.

list-accounts, get-connect-link

You in NanoClaw

Rewrite the draft you just saved so it opens with the question.

Bolta

I will change the draft text and leave it waiting. I will not approve, schedule or publish it.

Draft updated. Waits for your yes.

save-post

Bolta can draft for

  • LinkedIn
  • X
  • Instagram
  • Facebook
  • Threads
  • Bluesky

Publishing always follows your approval and your brand's posting rules.

When the chat is quiet

What keeps running while you are not messaging NanoClaw

NanoClaw is a process on your own hardware, so what continues depends on that machine staying on.

  • The agent group and its scheduled chat jobs keep running only while the host machine is up and its containers are running.
  • A scheduled job such as a Friday drafting request can only draft and send to review.
  • A Bolta routine runs on Bolta's side once you have confirmed it, even if NanoClaw is stopped. It stays switched off until you say yes to turning it on.
  • If the host is off, nothing on the NanoClaw side runs. Drafts already saved in Bolta stay there.

Unattended work stops at drafts and review. Publishing still needs your yes.

The agent

What NanoClaw is

NanoClaw is an open-source, MIT-licensed assistant you host yourself. You message it from a chat app and it runs your request. Each agent group runs in its own isolated containers.

For social media, that means you can say draft this week's posts from your phone and have the drafts appear in Bolta's review list. It is built for people who are comfortable with a terminal, since setup uses Docker and the ncl command line.

Credentials

Keep the Bolta key in the vault

NanoClaw separates settings the agent can read from secrets it cannot.

Keep the Bolta key in the vault
WhereSafe for the keyWhy
Credential vault (OneCLI Agent Vault)YesThe gateway injects the credential in flight, so the agent never holds it.
--headersNoMeant for non-secret header values.
--envNoNanoClaw's docs say not to put a raw key here.
The URLNoNanoClaw rejects URLs with credentials in them.
Always on

Chat apps and scheduled jobs

Because NanoClaw stays running on your machine, you can message it any time from a paired chat app such as WhatsApp, Telegram, Slack or Discord. You can also ask it to run a job on a schedule, for example a Friday request for next week's drafts.

A scheduled chat job can only do what the agent can already do. It drafts and sends to review. Publishing still needs your yes, and a repeating Bolta routine needs its own separate confirmation before it is switched on.

Agencies

One agent group per brand

Add Bolta to each group with its own ncl groups config add-mcp-server command. Because each group runs in its own containers with its own memory, a client's chat history stays in that client's group.

When you create a Bolta key, pick the one brand it is for (or tick only that brand) and give it only the permissions it needs. Add that key to that brand's group only.

Before Bolta

Install NanoClaw before you add anything

NanoClaw is cloned from github.com/nanocoai/nanoclaw and set up with bash nanoclaw.sh. It needs Docker, pnpm and Node.js, and the README holds the current versions.

Pair one chat app first and confirm you can message the agent. Adding Bolta to a group that you cannot reach from your phone makes it hard to tell which part failed.

Where to type it

Run ncl on the host, not inside the agent

The add-mcp-server command belongs on the machine where NanoClaw is installed. NanoClaw gates ncl write commands that come from inside an agent container, so running it from a chat message will be blocked.

Changes apply on the next container spawn. Running ncl groups restart --id <group-id> loads the new server straight away.

Daily use

Drafting from WhatsApp, Telegram, Slack or Discord

Write the request the way you would to an assistant, and name the brand. The agent saves a draft in Bolta and quotes the text back, so you can read it where you asked.

A yes in chat covers the one post you were shown. A different post, or a repeating routine, needs its own yes.

Safety

What NanoClaw cannot do without you

NanoClaw connects as: Remote MCP server added to an agent group with ncl.

  • Drafts are free, finals need your yes

    NanoClaw can write, edit and save drafts at any time. Approving, scheduling, publishing, sending a reply, activating a routine or campaign and deleting each wait for your yes, shown against the exact text.

  • One connection, one organization

    When you connect NanoClaw you tick the brands it may reach. It reaches only those, and when that is more than one brand, every call names the brand it is for.

  • Your role is the ceiling

    Each call NanoClaw makes is checked like a normal request from you. Owners and admins can act on any post, creators on Bolta's drafts and on posts they are a named reviewer on, and viewers never approve or publish.

Terms

The words behind NanoClaw and Bolta

For NanoClaw: Remote MCP server added to an agent group with ncl, signing in with API key held in NanoClaw's credential vault (OAuth not confirmed).

MCP
The Model Context Protocol, an open standard that lets an AI app call tools in another product. Bolta's hosted server is at https://mcp.bolta.ai/mcp, and NanoClaw reaches Bolta through a connection like it.
Skill
A folder of instructions an AI app reads so it knows how to use Bolta well. It does not connect by itself: install it with npx skills add boltaai/bolta-skills --skill bolta next to the connection.
Plugin
A bundle that brings the connection and the skill in one install, where the app supports plugins. Claude Code has one, added with its plugin marketplace commands.
API key
A bolta_sk_ key you create in Settings, sent as a Bearer header. It carries the permissions you choose, which suits apps you configure by hand or anything that runs unattended.
Read the MCP docs

Setup for NanoClaw is only partly confirmed. See what is and is not confirmed. Last checked 2026-11-21.

Not the same thing

NanoClaw is not OpenClaw

NanoClaw is its own open-source agent, with isolated containers per agent group. OpenClaw is a different self-hosted agent with its own setup, so the steps on this page do not apply to it.

FAQ

Frequently asked questions

Short answers about using NanoClaw with Bolta.

What is NanoClaw and is it free?
NanoClaw is an open-source, MIT-licensed assistant you install and host yourself. The software is free, though you pay for the hardware you run it on and for any model it uses. Bolta has its own plans and drafts are free.
How do I add Bolta to NanoClaw?
On the host machine, run ncl groups config add-mcp-server --id <group-id> --name bolta --url https://mcp.bolta.ai/mcp, then ncl groups restart --id <group-id>. Add your Bolta key through NanoClaw's credential vault, not in that command.
Do I need an API key or can I sign in with OAuth?
Plan on an API key from Settings, API Keys in Bolta. Whether NanoClaw's remote MCP servers support OAuth sign-in is not described in its docs, so this page does not promise it.
Where does my Bolta key go so the agent cannot read it?
In NanoClaw's credential vault (OneCLI Agent Vault by default), which injects it in flight. Never put it in --env or --headers: NanoClaw says --headers is for non-secret values and not to put a raw key in --env. The exact vault command is not confirmed here.
Can NanoClaw post to social media without my approval?
No. Bolta keeps drafts in review and refuses unapproved publishing. Safe Mode, roles and any client sign-off still apply, even when a request comes from a chat app.
Can I use a different NanoClaw agent group for each brand?
Yes, and it is a good fit for agencies. Add Bolta to each group separately, so each brand gets its own containers. Create one Bolta key per brand and add it to that brand's group only.
Which chat apps can I use to talk to NanoClaw?
NanoClaw's docs list WhatsApp, Telegram, Slack, Discord and Signal among others. You pair each one with an /add-<channel> setup. Check NanoClaw's docs for the current list.
Does NanoClaw work with the Bolta skill?
Not confirmed. NanoClaw documents skills as SKILL.md workflows, but an npx skills install is not described there, and Bolta's skill has not been tested in its containers. The hosted MCP server is the documented route.
How do I restart the agent after adding the server?
Run ncl groups restart --id <group-id> on the host. Otherwise the change applies the next time the container spawns.
Is this a verified setup?
Partly. The ncl command, restart and HTTPS rule come from NanoClaw's official docs. The vault step for a Bolta key and OAuth support have not been run end to end, which is why this page shows a caveat.
Does NanoClaw need to stay on for Bolta drafts to exist?
Only to create new ones through chat. Drafts already saved in Bolta stay there if the host is switched off. A Bolta routine you confirmed keeps running on Bolta's side.
Can I use NanoClaw without Docker?
NanoClaw's documented setup uses Docker, and each agent group runs in its own containers. This page does not cover a Docker-free setup because none is described in the docs it relies on.
Troubleshooting

If something does not work

Quick fixes for the common snags.

The agent says it has no Bolta tools
Changes apply on the next container spawn. Run ncl groups restart --id <group-id> and message the agent again. Also check you added the server to the same group you are chatting with.
ncl rejects the URL
NanoClaw needs an HTTPS URL with no credentials in it. Use exactly https://mcp.bolta.ai/mcp with no key or token in the address.
The server loads but Bolta calls fail with an authorization error
The key is probably not reaching Bolta. Check it starts with bolta_sk_, has the permissions you need, and that the vault is set to inject it for this host. This page cannot confirm the vault steps, so use NanoClaw's own vault docs.
The add-mcp-server command is blocked
NanoClaw gates ncl write commands that run from inside an agent container. Run the command from the host shell on the machine where NanoClaw is installed.
It drafts for the wrong brand
Ask it to list your workspaces first and name the brand in the request. For clients, use one agent group per brand so the wrong one cannot be picked.
The agent says it cannot publish
That is by design. Bolta refuses or sends to review anything you have not approved, and Safe Mode and role limits apply. Approve in Bolta or say yes after it shows the exact text.

Ready to schedule social media posts with NanoClaw?

Connect NanoClaw and ask for your first draft. Nothing goes out until you say yes.