AI client access

Turn every AI client into a scoped telecom agent.

Dialplane lets telcos connect Claude, ChatGPT, Copilot, Gemini, support chatbots, and automated MCP workers to BroadWorks without handing them raw XML, sockets, or unbounded credentials.

Request AI connector accessSee scoping modelSupport use cases
Remote MCP endpoint
HTTPS

https://mcp.dialplane.com/broadworks

Compatibility path: https://api.dialplane.com/mcp

Claude
ChatGPT / OpenAI
GitHub Copilot
Gemini

scope

5-layer stack

support

bots + copilots

savings

less L2 drag

One connection model

The setup is the same everywhere.

Create a scoped token

Issue a scoped token

Start read-only, then add exact scopes such as read:user, read:cdr, provision:user, batch:user, backup:*, or restore:*.

Add one MCP endpoint

Every supported AI client points at the same remote Dialplane endpoint over HTTPS.

Use named workflows

Agents call explicit tools instead of inventing raw OCI XML or opening BroadWorks sockets.

Verify every change

Dialplane normalizes responses and records before / after evidence for state-changing workflows.

What telcos can market

AI support that can inspect telecom state, not just summarize tickets.

Dialplane gives service providers a concrete AI story: customer chatbots, frontline copilots, and automated MCP agents that can safely read live BroadWorks state, prepare work, and hand off evidence.

Least privilege for every agent

Scoped telecom AI access

Expose only the read, provisioning, batch, backup, restore, or sync capabilities a workflow needs.

Answers grounded in live BroadWorks state

Support chatbots that can verify

Let customer-facing bots and internal copilots check users, groups, CDRs, and calling features without raw switch access.

Fewer manual lookups and escalations

Cost savings without risky shortcuts

Reduce repeated BroadWorks screen hops, ticket handoffs, and provisioning rework while keeping write actions bounded.

Before / after evidence

Audit-ready operations

State-changing workflows produce traceable records operators can review instead of relying on chat transcripts.

Telco support leverage

Show the exact ticket AI can handle before escalation.

A telco can run its own chatbot or agent framework, connect that agent to Dialplane over MCP, and let Dialplane enforce the customer, token, BroadWorks, and workflow boundaries on every call.

Example workflow

"Why are calls not forwarding?"

read-first
  1. 01Customer asks why calls are not forwardingThe telco chatbot collects the caller, user, number, and time window without exposing BroadWorks credentials to the conversation.
  2. 02Agent reads the live telecom recordDialplane lets the agent inspect user profile, CF Always, CF Busy, CF No Answer, schedules, group state, and recent CDR context through scoped read tools.
  3. 03Answer is grounded in normalized evidenceBroadWorks XML is returned as typed JSON, so the bot can explain the actual forwarding target, schedule conflict, disabled feature, or missed-call pattern.
  4. 04Escalation arrives already packagedIf a human needs to approve a fix, the ticket includes customer account, user, checked tools, before state, suspected cause, and the bounded workflow to run next.
Reference architecture
Telco AI layerlayer 1Customer chatbot, agent assist copilot, back-office worker, or your own MCP-aware automation.
Dialplane MCPlayer 2Remote HTTPS endpoint with OAuth or scoped tokens, customer isolation, and tool discovery.
Scoped telecom toolslayer 3Read, provision, batch, backup, restore, sync, and audit workflows with SP/group bounds.
BroadWorks evidencelayer 4Schema-rendered OCI commands return normalized JSON plus traceable before / after context.
Cost-savings proof points

Manual lookup time

One agent request can replace repeated BroadWorks screen hops across user profile, feature settings, schedules, groups, and CDR views.

Escalation load

Read-only bots can close simple status and routing questions, while complex cases arrive with the evidence L2 needs to decide quickly.

Provisioning rework

Named workflows validate inputs, execute the same steps every time, and verify after state before the operator calls the task complete.

Audit review

Before / after records give operations teams something stronger than a chat transcript when a customer or manager asks what changed.

Customer support chatbot

read:user read:cdr read:group

Answer routine questions like call forwarding state, missed-call patterns, voicemail status, and group membership before a rep opens BroadWorks.

Escalates with customer account, user, CDR window, and normalized evidence already attached.

Agent assist copilot

read:* backup:user

Give frontline reps a controlled assistant that can inspect current state, explain likely causes, and prepare a fix plan.

Keeps the human in the seat for approval while removing the lookup burden.

Automated MCP worker

read:* provision:user batch:user

Run bounded background agents for onboarding, number assignment, schedule checks, or spreadsheet-driven changes.

Uses named workflows instead of freeform XML and stays inside customer plus SP/group bounds.

Operations supervisor

read:* backup:* restore:*

Review batches, compare snapshots, watch scope denies, and restore known-good state when a controlled workflow needs recovery.

Turns AI activity into operational evidence instead of one-off chat history.

Automated MCP agent path
  1. 1Customer asks the telco chatbot or a rep asks their copilot.
  2. 2The telco-owned agent calls Dialplane over the branded MCP endpoint.
  3. 3Dialplane intersects customer account, token scope, SP/group bounds, login policy, and workflow guardrails.
  4. 4The agent returns an answer, drafts a change, or runs an approved workflow with evidence.

Access scoping

Every AI action has to pass every layer.

Dialplane does not trust an AI client just because it can connect. Customer isolation, token scopes, BroadWorks bounds, login ceilings, and workflow guardrails all apply at the same time. The narrowest permission wins.

Permission stack
deny by default
  1. 01Customer boundarycustomer accountEvery request resolves to one customer environment before any tool runs. Service providers and groups are BroadWorks dimensions inside that customer boundary.
  2. 02Token scoperead:user + backup:*Each token carries explicit scopes. A read token cannot provision, a CDR token cannot manage users, and a workflow token only exposes the verbs it was granted.
  3. 03BroadWorks boundsSP / groupTokens can be constrained to specific service providers or groups, so an AI client for one team cannot wander across the whole customer environment.
  4. 04Login-type ceilinguser < group < SPBroadWorks credentials still matter. Dialplane intersects token scope with the login level for user, group, service provider, reseller, provisioning, or system sessions.
  5. 05Workflow guardrailsconfirm + auditMutating tools use named workflows, schema-rendered XML, destructive-operation bans, before / after reads, redaction, and audit records.

Most restrictive side wins

allowed = customer + token scope + BroadWorks bounds + workflow guardrails

Read-only AI assistant

Bound to one service provider or support group

read:user read:group read:cdr

Can

Answer inventory, CDR, and user-state questions with normalized JSON.

Cannot

Provision users, change calling services, restore backups, or upload CDRs.

Provisioning operator

Bound to the groups that team manages

read:user provision:user provision:dn batch:user

Can

Create bounded users, assign numbers, run spreadsheet-style batches, and verify after state.

Cannot

Touch other groups, delete account-level containers, or bypass BroadWorks login ceilings.

Backup and restore reviewer

Bound to selected service providers or groups

read:* backup:* restore:*

Can

Capture snapshots, compare changes, and restore known-good state through audited workflows.

Cannot

Freestyle raw XML, read unrelated customer environments, or restore outside the granted bounds.

CDR analyst

Customer-scoped analytics only

read:cdr

Can

Ask for missed calls, trunk demand, usage trends, and group activity.

Cannot

Upload CDR files, change BroadWorks configuration, or call admin APIs.

This is the difference between “the model has telecom access” and “the model has exactly the telecom access this workflow needs.”

Read security details

Client-specific paths

Claude, ChatGPT, Copilot, and Gemini are already MCP-shaped.

Dialplane exposes the server. Your AI client handles discovery and confirmation UX; Dialplane enforces customer, scope, bounds, and workflow policy on the server.

Custom connector

Claude

Add Dialplane as a remote MCP connector so Claude can discover scoped BroadWorks tools for support, operations, and engineering workflows.

  1. 1Add a custom Web connector in Claude settings.
  2. 2Enter the Dialplane MCP endpoint.
  3. 3Authenticate with OAuth or a scoped Dialplane token.
  4. 4Enable the connector only in conversations that need telecom access.
Connector URL
https://mcp.dialplane.com/broadworks

Apps, connectors, API

ChatGPT / OpenAI

Use Dialplane as a remote MCP server for ChatGPT apps, support copilots, deep research, or API workflows that need controlled telecom tools.

  1. 1Create an app or connector that points at the Dialplane MCP server.
  2. 2Use OAuth for user-level access when publishing to a workspace.
  3. 3Allowlist the read or workflow tools the assistant should use.
  4. 4Require confirmation before state-changing telecom actions.
{
  "type": "mcp",
  "server_label": "dialplane",
  "server_url": "https://mcp.dialplane.com/broadworks"
}

VS Code or Copilot CLI

GitHub Copilot

Connect Copilot to Dialplane through an HTTP MCP server entry for implementation work, audits, and support automation prompts.

  1. 1Add Dialplane to your MCP server configuration.
  2. 2Use HTTP / Streamable HTTP transport.
  3. 3Pass a scoped API key or OAuth bearer token.
  4. 4Ask Copilot to use Dialplane when reviewing or changing telecom state.
{
  "mcpServers": {
    "dialplane": {
      "type": "http",
      "url": "https://mcp.dialplane.com/broadworks",
      "headers": { "X-API-Key": "bwb_sk_..." }
    }
  }
}

Gemini CLI

Gemini

Register Dialplane as an HTTP MCP server so Gemini CLI can call customer-safe BroadWorks workflows from the terminal or from automated agent runs.

  1. 1Add an HTTP MCP server named dialplane.
  2. 2Attach the scoped token as a header.
  3. 3Inspect discovered tools with the MCP status commands.
  4. 4Keep confirmation prompts on for mutating workflows.
gemini mcp add --transport http \
  --header "X-API-Key: bwb_sk_..." \
  dialplane https://mcp.dialplane.com/broadworks

Built for delegated operations

Give agents tools, not the keys to the switch.

Dialplane sits between AI clients and BroadWorks with customer isolation, schema validation, redaction, and audit trails. Teams can start with read-only support answers, then add provisioning and batch scopes when the workflow is ready.

Scoped tokens and OAuth sessions keep customer environments separated.

Tool scopes and BroadWorks service provider or group bounds limit what each AI client can reach.

Chatbot confirmations happen in the client; Dialplane still enforces server-side scope and bounds.

Destructive account-level deletes are blocked across MCP, REST, and raw OCI passthrough.

BroadWorks XML is rendered from schemas and returned as typed, camelCase JSON.

Buyer questions

The short version for security, support, and operations.

Dialplane is built so a telco can market AI support without promising unbounded automation or giving a model direct control of the switch.

Does the AI get BroadWorks credentials?

No. The AI client calls Dialplane. Dialplane holds the connection policy, applies scopes, and returns normalized evidence instead of exposing raw switch access.

Can an AI client make changes?

Only when the token, BroadWorks bounds, login level, and named workflow all allow it. Read-only agents can inspect state but cannot provision, restore, or batch changes.

How do scopes work?

Start with read scopes such as read:user or read:cdr, then add workflow scopes like provision:user, batch:user, backup:*, or restore:* only for approved use cases.

Can this support our own chatbot?

Yes. Your chatbot, copilot, or automation worker can call Dialplane over remote MCP while Dialplane enforces customer, scope, and BroadWorks boundaries server-side.

Can the endpoint be branded?

Yes. Telcos can put their own AI support layer in front while Dialplane provides the controlled MCP bridge, tool discovery, workflow execution, and evidence records.

Next step

Put Dialplane behind your AI support layer.

Tell us which AI client, chatbot, or automation framework you want to connect. We will map the first read-only scope set and the safest workflow path.

Request AI connector accessReview client setup