AI CRM Nepal: APIs, Webhooks and MCP for AI Assistants

A modern CRM should participate in the work around it

A customer relationship rarely lives inside one application. A Nepali business might receive an enquiry through a website, discuss it on the phone, assign a salesperson, deliver a project and pass selected information to another system. When those steps require repeated copying, the CRM becomes one more place to update instead of a useful working centre.

AI CRM Nepal addresses that problem through three concrete integration surfaces: APIs, outgoing webhooks and Model Context Protocol tools. They connect ordinary CRM records with external programs, event-driven workflows and AI assistants. These are implemented interfaces with public references, not a promise that a future AI feature will arrive someday.

This article explains what each interface supports, which assistant connection routes are documented, and how to turn a business instruction into an authorized CRM action. The counts in this guide reflect the public API and MCP references reviewed on 20 October 2026. Your own connection may see fewer tools because its permissions are narrower.

Official references: Public API reference · Live MCP tool catalog

API-friendly, webhook-friendly and MCP-friendly mean different things

An API gives a program an agreed way to request an operation. Your server supplies the authentication and required fields; the CRM validates the request and returns a result. This is useful when the operation is predictable, such as creating a lead from a separate application or adding the outcome of a call.

A webhook works in the other direction. After a configured event, the CRM sends information to your receiver. That receiver can start another workflow, update an external system or put work in a queue. The receiving application must still decide what to do and confirm its own processing.

MCP gives a compatible assistant a structured tool interface. The client can discover available tools, inspect their parameters and invoke approved actions in response to your instruction. AI CRM Nepal supplies the CRM capabilities; your chosen AI client interprets the instruction and decides which permitted calls to propose or execute. Client approval settings still apply.

Official references: MCP introduction

The six public automation APIs available today

AI CRM Nepal publishes six external automation endpoints. All use POST with JSON, including the lead-details read. Configure the workspace's Automation settings and keep its Bearer token in your backend's secret storage. Use the API reference and machine-readable OpenAPI file for the exact fields, validation rules and error responses.

These operations give a developer a practical customer workflow: capture the initial enquiry, retrieve the existing relationship, write the outcome, move the sales stage, assign responsibility and create a concrete task. Choose the endpoint that represents the intended change rather than relying on a browser-only route.

• POST /api/automation/intake — create a lead with a name and phone, supported optional details and generated CRM identifiers. An existing phone returns a duplicate conflict; this endpoint is not an automatic upsert.
• POST /api/automation/lead-details — look up a lead and optionally retrieve its notes. Supported opt-ins also expose form-submission and WhatsApp-enquiry context with their own pagination.
• POST /api/automation/lead-note — append a note to a recognized lead, such as a conversation summary or an agreed next step.
• POST /api/automation/lead-pipeline — change the lead's stage using the supported stage value or configured option reference.
• POST /api/automation/lead-assignee — change responsibility using the documented active workspace-member assignment fields.
• POST /api/automation/task-create — create a task, optionally connected to a lead, with the documented title, assignment and scheduling fields.

Official references: Endpoint fields and examples · OpenAPI specification

Stable customer references make integrations more useful

A useful integration needs to recognize the same customer when a later request arrives. AI CRM Nepal supports lead references through a CRM ID, Supabase UUID, internal Accounting ID or phone number in the documented lookup and update operations. Keep the returned identifiers in the external system's mapping instead of repeatedly matching names.

For example, an external enquiry system can store the CRM ID returned when it creates a lead. A later call outcome can use that ID to add a note to the intended relationship. The internal ACC identifier is another stable CRM reference; mapping it to a separate accounting provider's customer ID remains part of the integration you build.

History also matters. An assistant preparing a callback can request permitted notes and continue through the available pages. Form answers and WordPress WhatsApp enquiries can provide additional customer intent when explicitly requested and authorized. Reading every relevant page gives the assistant a better basis for a useful summary than a single latest-note snippet.

Official references: Lead identifiers and paginated context · MCP context and parameter reference

The four outgoing webhook events

In Automation, configure the publicly reachable HTTPS destination and enable the events your workflow needs. Separate event settings let a business route new enquiries, sales progression, ownership changes and due follow-ups to the appropriate receiving workflows.

The event carries its name and workspace/lead context. Stage and assignment changes provide transition information; due follow-ups include scheduling and event identity information. Your receiver can use those details to decide whether to create an external work item, notify an authorized team member or start a separately configured agent workflow.

• lead.created — a lead-creation event for a configured intake or workflow receiver.
• lead.pipeline_changed — a stage transition, including previous and new stage information.
• lead.assignee_changed — a change in lead responsibility; the documented default-only assignment case is skipped.
• lead.follow_up_due — a scheduled follow-up event handled through the due-event dispatch worker.

Official references: Webhook events, payloads and delivery

Build a webhook workflow that can confirm its result

An example is a lead moving to a stage that requires a technical review. The configured stage-change webhook reaches your n8n webhook receiver. Your workflow checks the event, identifies the lead and chooses the next action. It might request a CRM task through the public API or pass a bounded instruction to an agent connected through MCP.

AI CRM Nepal documents timestamped HMAC-SHA256 signatures and event/workspace headers for receivers to verify. Verify the exact raw body before processing, reject an unexpected event or workspace, and follow the documented secret handling. Return a prompt success response after accepting work, then handle longer processing in your own queue or workflow.

Plan for retries and repeated delivery. Use the documented event identifiers where applicable and an appropriate duplicate-handling strategy. The ordinary API note/task writes are not universally idempotent, so reconcile an uncertain result before retrying them. A due-follow-up workflow depends on its dispatcher running; a successful CRM write alone does not prove the receiver completed an external action.

Official references: Signature verification and retry guidance

MCP turns an assistant into a permitted CRM operator

AI CRM Nepal's remote MCP endpoint is https://www.aicrmnepal.com/api/mcp, using Streamable HTTP. Its published catalog currently contains 51 tools. The server supports OAuth with PKCE and personal Bearer-token connections, depending on the client. Your personal MCP connection and the separate workspace automation API token serve different interfaces and should be configured accordingly.

The assistant can begin with crm_identity to understand the acting workspace and member, then crm_schema to discover permitted tools and their input requirements. Tool descriptions include required and optional parameters, limits and synthetic examples. This gives the client a concrete contract for taking an action instead of guessing a button, database field or private endpoint.

The important result is practical: when you authorize a compatible assistant and grant the relevant capabilities, it can read records and submit supported changes to the CRM. It can do work on your behalf through tools. This capability is broader than an AI-generated explanation of how you might manually update a record.

Official references: Connection and authentication guide · All 51 tool definitions

What the MCP tools cover

The MCP catalog covers customer records, operational work and reporting, with separate permissions for different responsibilities. The full catalog is a description of what the server implements, not an automatic grant of every action to every person or AI connection.

This breadth lets a sales assistant work across a relationship: understand an enquiry, record the next step, coordinate delivery work and explain progress from authorized reports. An administrative assistant can receive a different, deliberately approved set of responsibilities. The workspace can keep financial data and high-impact changes outside a sales assistant's scope.

• Identity and discovery: crm_identity and crm_schema identify effective access and explain the available tool parameters.
• Leads and customer context: list/get/create/update leads; read notes and activities; append activity; assign members; change stages and follow-up times.
• Lead transfer and recovery: bounded import/export, deletion history and snapshot restore, each requiring the appropriate authority. Snapshot restore is not a promise to recreate every old activity record.
• Revenue: separately permissioned earned-total and adjustment-ledger reads/updates.
• Tasks: list/get/create/update/delete tasks and read/add task comments.
• Projects: list/get/create/update/delete projects, inspect project-lead relationships and read/add project comments.
• Reporting: dashboard, analytics, team performance, usage, plan and audit tools.
• Workspace administration: permitted business-profile, ordinary workspace-setting and custom-option reads/updates; separately delegated ordinary-member invitation/access and MCP member-permission tools within existing authority ceilings.

Official references: Exact tools, scopes and schemas

Which AI assistants and agent clients can connect?

AI CRM Nepal documents connection approaches for ChatGPT, Claude, Claude Code, Codex, n8n and Gemini CLI. Choose a client that supports this remote MCP transport and the appropriate authentication. Availability can depend on the AI provider's plan, account controls and current client version. The setup guide is the place for the current connection recipe.

• ChatGPT web: use a custom MCP app with OAuth where the account and workspace allow it. The assistant can use the tools granted to that connection; eligible app/developer-mode access is controlled by OpenAI.
• Claude web and Claude Desktop: use a remote custom connector with an appropriate sign-in configuration. This is the remote-connector route, rather than a claim that a local desktop extension is already installed.
• Claude Code: configure the CRM as a remote HTTP MCP server and authenticate through the client's supported OAuth or header flow.
• Codex: configure a Streamable HTTP MCP server, with OAuth or a Bearer token sourced from an environment variable according to the client setup.
• n8n: connect an MCP Client Tool to an AI Agent and configure the CRM endpoint and authentication. For a fixed operation, an ordinary HTTP Request step against the automation API can be the clearer choice.
• Gemini CLI: configure an HTTP MCP server using httpUrl and supported authentication headers. This specifically describes Gemini CLI, rather than every consumer Gemini interface.

Official references: AI CRM Nepal client setup · OpenAI ChatGPT MCP apps · Claude remote connectors · Claude Code MCP · Codex MCP · n8n MCP Client Tool · Gemini CLI MCP

What about Grok and custom AI agents?

A developer-built agent can use the six public automation APIs through its backend, regardless of which model interprets the user's request. That backend must translate the instruction into a supported operation, validate its inputs and keep CRM credentials away from public browser code. The model name alone does not create a CRM connection.

xAI also documents remote MCP tools with HTTP/SSE transport, authorization headers and tool filtering. This provides a developer route to evaluate for a Grok-powered application. A complete AI CRM Nepal–xAI execution flow has not been independently verified for this article, and this is not a claim that the consumer Grok chat interface has a ready-made CRM connector. Validate the actual client, token handling, schema and approvals before granting write access.

Official references: xAI remote MCP developer documentation · CRM external API contract

What can you ask an authorized assistant to do?

Start with a specific business outcome and the intended record. An instruction such as 'Find the lead using this CRM ID, read its permitted history and summarize the agreed next step' maps to discovery, lead lookup and context tools. The assistant should complete the relevant note pages before presenting the history as complete.

A follow-up instruction could be 'Add a note that the customer requested a demonstration, set the next follow-up to tomorrow at 11:00 Nepal time and create a preparation task for the assigned salesperson.' With the necessary note, follow-up and task capabilities, the assistant can submit those supported changes. If the requested assignee or option is unclear, it should discover or ask rather than invent one.

For delivery work, you might ask it to review an authorized customer project, add an update comment and create the next task. For a team review, request a summary based on permitted dashboard or performance tools. These are illustrative instructions, not claims about a measured result or a particular customer's deployment.

Successful control means the CRM accepted the intended operation and the assistant reports that result accurately. If a write returns only mutation metadata, it should perform a separately permitted read when record content is needed. An assistant should not report a customer message, payment, phone call or external workflow as completed merely because a CRM task or note was saved.

Your team chooses how much control the assistant receives

MCP access combines workspace policy, an enabled member grant, connection consent and ordinary CRM authorization. A salesperson can have a different tool set from an owner. Assigned-only visibility and separate permissions for notes, financial information, exports and destructive changes continue to matter when the action comes from an AI tool.

Begin with a narrow useful role. A research assistant may need lead/context reads and no writes. A sales assistant may also need note creation, follow-up scheduling and task creation. Team-management or permission-delegation tools need their own explicit authority; an assistant does not gain administrator status simply because it can call another CRM tool.

The user can revoke a connection, and dispatch rechecks current effective authority. MCP mutations use stable idempotency keys for safe identical retries; supported version checks help avoid overwriting concurrent changes. The client can impose additional confirmations. These controls let a business introduce agent actions deliberately instead of sharing an unrestricted owner credential.

Official references: Permissions, retries and connection lifecycle

An example workflow for a Nepali service business

Imagine a service company collecting website enquiries. A visitor submits a CRM form with a phone number and a question. Staff can find the relationship in the CRM, and an appropriately authorized assistant can read the relevant submitted context. A WordPress site can also use the separate capture flow to save a phone enquiry before opening the configured WhatsApp destination.

The salesperson asks the assistant to prepare the conversation from the available record. After a human call, the assistant saves the factual outcome and schedules the next step. Moving the lead to a configured stage can send a webhook to the company's receiver, which starts a separately configured workflow. The assistant can then help with authorized tasks and project handoffs as the relationship develops.

Each part has a clear responsibility: forms capture the enquiry, APIs connect fixed application steps, webhooks notify a receiver, and MCP lets a compatible assistant operate the permitted CRM tools. The example does not assume that a CRM subscription itself supplies a telephony account, an external messaging service or every receiving workflow.

Official references: Enquiry-to-delivery product walkthrough · WordPress capture flow

How to start with one real business workflow

Choose one task that currently causes repeated copying or an unclear handoff. Decide whether it is a fixed API operation, an event receiver or an assistant-led sequence. Agree what success looks like before connecting the whole process: one correctly identified lead, one meaningful note and one assigned next step is a useful first milestone.

For APIs, follow the Automation setup and public endpoint reference. For webhooks, prepare and verify the receiver. For MCP, configure workspace/member permissions, connect the chosen client and inspect effective discovery before testing. Use fictional information first; confirm saved results and the receiving system's outcome separately.

Then expand only the responsibilities the team actually needs. Include CRM plan limits, AI-client access and external provider/workflow costs in the operating plan. AI CRM Nepal provides the integration surfaces; the connected system and the authorized workflow determine how far the overall automation goes.

Official references: API setup · MCP setup · CRM plans

Why this is a practical definition of a modern CRM in Nepal

A modern CRM should help a team work with customer information through more than a manual screen. AI CRM Nepal combines its ordinary lead, task and project workspace with documented APIs, configured event delivery and a permission-controlled tool interface for compatible assistants. Those are concrete reasons to describe it as API-friendly, webhook-friendly and MCP-friendly.

The strongest demonstration is the work a business can perform: a program creates an enquiry, a receiver reacts to a sales event, and an authorized assistant finds context and submits a permitted next step. The product's public references make those capabilities inspectable. Evaluate the actual workflow and its results rather than assuming that an AI label means every business process runs on its own.

Common questions

Sources and review notes

Published by AI Automation Nepal Pvt. Ltd., the maker of AI CRM Nepal, with AI-assisted drafting. Reviewed 20 October 2026 against current source, live public OpenAPI/MCP references and linked client documentation. Example workflows are illustrative. Client setup recipes are not fresh end-to-end acceptance tests in every provider account; the xAI route is explicitly unverified with this CRM. Capabilities, provider plans and configuration requirements can change. No independent market-share, performance or universal compatibility claim is made.

• AI CRM Nepal API documentation
• AI CRM Nepal OpenAPI contract
• AI CRM Nepal MCP connection guide
• AI CRM Nepal complete MCP catalog