Claude + Codex MCP · Private preview

Connect Claude and Codex to your Polxy workspace.

Use Polxy directly from Claude, Claude Code, or Codex. Ask about projects, campaigns, Queue work, connections, credits, and reports—prepare guarded drafts, generate images, or start missions after reviewing their credit cost—with your workspace permissions.

  • Read, create + missions
  • Workspace-scoped
  • Reviewed generation
Secure context bridge Guarded mode
Supported clients CLAUDE + CODEX Ask in natural language
Authorized workspace POLXY Permission-scoped context
Projects 12 Campaigns 28 Queue 9 Reports 16

Claude or Codex receives only the data and tool permissions granted to your authenticated Polxy role.

Current build

The Polxy MCP server and OAuth 2.1 sign-in flow are available to Claude and Codex in private preview. Remote discovery, OAuth, tool loading, read calls, and approval-gated actions have local protocol and fixture coverage. Connected-provider workflows require their own acceptance checks.

View rollout status
One connection, familiar questions

How Polxy MCP works

MCP is the secure bridge between Claude or Codex and Polxy. For every request, Polxy verifies the user, workspace, permission, and exact tool before returning a bounded response.

  1. 01

    Add the connector

    Add the hosted Polxy endpoint in Claude connector settings, Claude Code, or Codex once private-preview access is enabled.

  2. 02

    Authorize Polxy

    Sign in through Polxy, approve the requested access, and select the workspace your role is permitted to use.

  3. 03

    Ask Claude or Codex

    Use normal language: “What is waiting for approval?” or “Create a LinkedIn draft for the Acme campaign.”

  4. 04

    Review the result

    Polxy returns workspace context, saves drafts for review, or generates an image after you approve its exact credit cost. Finished images appear in Studio and your conversation.

Interactive example

See the conversation before you connect

Choose a sample question to see which Polxy tools Claude or Codex would call and what a safe answer could look like.

Try a question
Example conversation Guarded-action preview
You

What needs my attention today?

Polxy tools used get_workspace_summary get_queue_status
AI

Your Queue has 9 items. Three are waiting for approval, including the Q3 launch video scheduled for tomorrow. Two connection checks also need attention before the remaining posts can be prepared.

I only inspected your workspace. Nothing was approved, scheduled, published, or changed.

39 permission-scoped tools

What Claude and Codex can do

Read workspace context, prepare campaign drafts and review-stage Queue work, or create images, start missions, and run SEO/GEO audits and prepare fix packages with separate permissions and exact cost reviews.

get_workspace_summary

Plan, active project, workload counts, credits, and automation state.

list_projects

Authorized projects with safe summaries and activity counts.

get_project_context

Audience, offer, goals, brand voice, and project workload.

list_campaigns

Campaign goals, channels, status, KPIs, and schedule context.

get_queue_status

Approval pressure, schedules, receipt states, and pending work.

get_credit_status

Credit balance, plan capacity, billing period, and safe ledger entries.

check_connections

Connection readiness without opening OAuth or contacting providers.

list_reports

Report summaries, measured status, proof counts, and timestamps.

create_campaign_draft

Create a non-active campaign draft without generating, scheduling, or spending.

create_content_draft

Prepare copy or a media brief as a Draft outside executable Queue states.

add_to_queue

Move an MCP-created draft into Needs approval—never directly to execution.

request_approval

Ask a human to review eligible work without granting approval.

run_campaign_preflight

Check drafting and Queue readiness while execution remains blocked.

get_action_status

Read the durable Polxy resource and receipt for an earlier MCP action.

cancel_queued_action

Cancel Draft or review-stage work; approved and executable states are blocked.

preview_image_generation

Review one image, its provider and exact credit cost. The review expires after 15 minutes.

generate_image

Create the reviewed image after your approval. Charges the quoted credits on success and saves it in Studio.

get_image_generation

Check the saved receipt after completion or a timeout. Retrying the same job does not create another image.

get_media_asset

View a generated image directly in your AI conversation. The full image remains in Studio.

list_missions

Find your project’s saved missions and available workflow types.

preview_mission

Review the exact target, channels, schedule, workload and preparation cost before starting.

start_mission

Start the reviewed mission and its internal work once. Public actions and media generation still require separate approval.

get_mission_status

See actual task and Queue progress, blockers and recorded preparation credits.

pause_mission

Stop future mission work. Actions already in progress may still finish.

list_search_audits

List saved SEO/GEO audits in your chosen project.

preview_search_audit

Review a free SEO crawl or a 640-credit GEO audit, its public target, provider context and read limits.

run_search_audit

Run the approved audit and save its report. No website edits or indexing submissions.

get_search_audit_status

Check the saved receipt after completion or interruption without repeating the audit.

get_search_audit

Read findings, page evidence and answer samples. Readiness estimates are separate from measured results.

list_search_fix_packages

List fix-package reviews and receipts created through MCP.

preview_search_fixes

Review exact draft contents from a saved audit: SEO costs 300 or 380 credits; GEO costs 260 credits.

create_search_fixes

Save the approved package once. SEO enters Human review; GEO remains draft-only. No website changes.

get_search_fix_package

Read package status and current draft bodies without creating or charging again.

list_feature_operations

Find named operations across Polxy modules, including native tools and browser handoffs.

describe_feature_operation

Inspect inputs, examples, preflights and effects before choosing an operation.

read_feature_records

Read saved project records for ads, pages, creators, outreach, affiliate, community, Reach, Telegram and search.

preview_feature_operation

Review one named operation, its exact inputs, selected project and maximum request credits.

execute_feature_operation

Run the approved operation using your current role and the app’s existing safeguards. Requires additional feature-execution consent.

get_feature_operation

Retrieve the saved result after completion or interruption. Uncertain work is never automatically repeated.

Choose your client

Choose where you use Claude or Codex

The endpoint is the same, but each client has a different setup screen. Follow only the card that matches where you are working.

Claude web · Recommended for Claude chat users

Connect inside Claude settings

Open Claude on the web, then add the Polxy URL as a custom connector:

https://www.polxy.ai/mcp
Do not paste this URL or a terminal command into a Claude chat. Chat messages cannot install connectors.
  1. 1

    Go to Settings → Connectors → Add custom connector.

  2. 2

    Name it Polxy, paste the URL above, and select Add.

  3. 3

    Select Connect, complete the Polxy browser sign-in, then enable Polxy under Search and tools in your chat.

On Team or Enterprise, an owner may need to add the connector for the organization before members can connect their accounts. Claude connector instructions

Claude Code · Local terminal only

Connect from Claude Code

Run this command in macOS Terminal or another local shell where Claude Code is installed:

claude mcp add --transport http polxy https://www.polxy.ai/mcp
Terminal command: do not paste it into Claude web chat. A hosted chat cannot access the local claude command.
  1. 1

    Start Claude Code locally and enter /mcp.

  2. 2

    Select Polxy and complete the secure browser sign-in.

  3. 3

    Confirm Polxy shows as connected. Check later with claude mcp list.

Codex · Desktop app, CLI, or IDE

Add Polxy to Codex

In the ChatGPT desktop app with Codex, open Settings → MCP servers → Add server. Choose Streamable HTTP and use:

https://www.polxy.ai/mcp
Add this in MCP server settings, not in a Codex chat message.
  1. 1

    Name the server Polxy, save it, then select Restart.

  2. 2

    Select Authenticate and complete the Polxy browser sign-in.

  3. 3

    Type /mcp in the composer to confirm Polxy is connected and its tools are available.

The desktop app, Codex CLI, and IDE extension share the same MCP configuration on one Codex host. Codex MCP instructions

Designed around boundaries

Your AI connection does not become a master key.

Every request is checked against an authenticated Polxy identity, one exact workspace, and the specific read, content, Queue, image, mission, search-audit, search-fixes, or additional feature-execution permission required by that tool.

Read about Polxy security

Permission scoped

OAuth access follows the current Polxy role. Draft, Queue, image, mission, search-audit, and search-fixes tools require their own scopes. Existing connections need new consent before generating images, starting missions, running SEO/GEO audits, creating fix packages, or operating additional app features.

Tenant isolated

The authorized workspace must exactly match the workspace being requested.

Sensitive fields removed

Provider credentials and authentication secrets stay out. Authorized business records may include the contact details needed for reviewed work.

Execution stays gated

Additional feature execution requires separate consent and approval of the operation and credit ceiling. Native publication proofs and provider readiness still apply. Sign-in, credential entry and payment authorization use browser handoffs.

Private-preview setup

What setup will look like

Once your workspace is invited, Claude and Codex both connect to this hosted endpoint. Polxy handles sign-in and authorization—never paste a normal Polxy browser session or installation command into a chat.

Private-preview connector URL · Add in connector settings https://www.polxy.ai/mcp

  1. 1
    Choose the correct setup

    Claude chat uses Settings → Connectors. Claude Code uses a local terminal. Codex uses MCP server settings.

  2. 2
    Add the Polxy URL

    Use the hosted MCP endpoint supplied with your preview invite.

  3. 3
    Sign in and consent

    Authorize one Polxy workspace through the secure browser flow.

  4. 4
    Start asking

    Claude or Codex can read context, prepare drafts, generate reviewed images, and start or pause missions when your role and granted permissions allow it.

Questions, answered

MCP connector FAQ

Which clients can connect?

Claude custom connectors, Claude Code, and Codex are the supported setup paths in this private preview. They connect to the same hosted Polxy MCP endpoint and authorize one permitted workspace through OAuth 2.1.

Why did the command fail in Claude or Codex chat?

A chat message cannot install an MCP connector or run a command on your computer. In Claude web, add Polxy under Settings → Connectors. In Claude Code, run the command in a local terminal. In Codex, use Settings → MCP servers.

Can the AI publish posts or spend ad budget?

With additional feature-execution consent, the client can request the same reviewed delivery and website operations available to your role in Polxy. Existing native approvals, account readiness and provider restrictions still apply. Older read and draft grants cannot approve posts, authorize public campaign delivery, publish, send, or spend ad budget. Provider charges, ad budgets and later scheduled runs must be reviewed separately from the request credit ceiling. Browser sign-in and payment authorization remain required. Catalog availability does not certify every connected-provider workflow.

Can the AI see every Polxy workspace?

No. Authorization is bound to one authenticated identity and exact workspace. The user’s Polxy permissions still apply.

Will this work with multiple businesses later?

The connector is already workspace-scoped. A future master-account feature could let an authorized user choose among assigned business workspaces while preserving isolation between them.

What happens if a user loses workspace access?

Every token use rechecks active membership and the user’s current role. Access tokens expire, refresh tokens rotate, authorizations can be revoked, and removed members immediately lose MCP access.

Polxy MCP private preview

Give Claude or Codex the Polxy context your team already has.

Request preview access