How to Set Up an MCP Server for Vendor Risk Monitoring in Claude, ChatGPT and Cursor (2026)

Tags:
  • Third Party Risk Management

Step-by-step guide to configuring the Bitsight MCP server in Claude, ChatGPT, and Cursor so your AI assistant can monitor vendor risk in plain language.

Third-party risk, governance, risk and compliance (GRC), and security operations teams increasingly want their AI assistant to pull live vendor cyber risk data on demand, instead of toggling between dashboards and spreadsheets. The Model Context Protocol (MCP) is the open standard that makes that possible. Our hosted MCP server at https://mcp.bitsighttech.com/mcp connects Claude, ChatGPT, Cursor, VS Code, Kiro, and Microsoft Copilot to seven core MCP tools and more than 1,000 Bitsight API operations covering ratings, findings, alerts, vendor portfolios and threat intelligence. This guide walks through what MCP is, why it matters in 2026, step-by-step setup for each client, continuous vendor monitoring workflows, permissions and data handling, and answers to the questions we hear most often.

What is an MCP server for vendor risk monitoring?

The Model Context Protocol (MCP) is an open standard developed by Anthropic and introduced on November 25, 2024, that gives AI assistants a consistent way to reach outside tools and data. The protocol separates three roles: the MCP host (the application the user interacts with, such as Claude Desktop or Cursor), the MCP client (the connector inside that host), and the MCP server (the external system that exposes data and actions).

An MCP server for vendor risk monitoring exposes vendor ratings, findings, risk vectors, vendor intelligence, and threat intelligence to an AI assistant so a third-party risk or GRC analyst can ask questions in plain language instead of writing API calls. The Bitsight MCP is our own server and a leading option in this category: it gives an MCP-compatible assistant, such as Claude, ChatGPT, or Microsoft Copilot, direct access to your Bitsight data and the full Bitsight API. It exposes seven core tools, covering API discovery, schema inspection, Knowledge Base search, and read and write operations, backed by more than 1,000 Bitsight API operations across ratings, findings, risk vectors, vendor intelligence, vulnerability context, and threat intelligence.

Why MCP servers for vendor risk monitoring matter in 2026

Vendor alert volumes keep climbing, and risk and security teams cannot keep pace through manual triage alone. Our MCP server and agentic capabilities close that gap by letting AI agents pull Bitsight ratings, findings and alerts at machine speed from a single intelligence platform, bridging the long-standing divide between risk and security work. An MCP-enabled environment now resembles a dynamic supply chain: you trust the model, the host application, every MCP server it reaches, the tools those servers expose and the credentials behind them. That is why an authenticated, vendor-backed MCP matters.

Common challenges in vendor risk monitoring and how an MCP server solves them

Third-party risk and GRC teams tell us the same story across onboarding, continuous monitoring, and offboarding: portfolios keep growing, alerts keep multiplying, and analyst headcount does not. Our Vendor Risk Management product exists to take the complexity out of managing that lifecycle, and Continuous Monitoring layers on objective data with a proven correlation to security incidents. The Bitsight MCP server extends both into the AI tools analysts already use, so a question like "which vendors declined this week?" returns an answer in seconds instead of a CSV export and a spreadsheet.

Key Problems Encountered

  • Fragmented alerts across a growing vendor portfolio. As the number of monitored vendors climbs, teams lose the ability to see which alerts actually matter across onboarding, ongoing oversight, and offboarding stages.
  • Manual API and CSV exports slow triage. Pulling ratings, findings, and alerts by hand through API scripts or portal exports adds hours between a risk signal and a decision.
  • Analyst capacity does not scale with alert volume. Alert surges from new vendors, emerging CVEs, and underground activity outpace the number of reviewers available to read them.
  • Inconsistent human-in-the-loop handoffs for vendor outreach. Without a shared workflow, escalations to vendor owners and procurement vary by analyst and get lost between tools.

The Bitsight MCP server addresses each of these directly. Vendor Risk Management handles portfolio-wide lifecycle complexity, Continuous Monitoring supplies the underlying signal, and the MCP lets AI agents and agentic GRC workflows consume that intelligence at machine speed without human-in-the-loop data retrieval. Because third-party risk management, exposure management, continuous monitoring and cyber threat intelligence sit on one Bitsight platform, analysts can triage, prioritize, and route vendor issues from a single conversation with their assistant and keep the final outreach decision with a human.

What to look for in an MCP server for vendor risk monitoring

Before you wire an AI assistant into your third-party risk program, evaluate any Model Context Protocol (MCP) server against a short list of operational, security and coverage criteria. The right server should be reachable from the assistants your team already uses, respect the permissions you have already set in your cyber risk platform, cover the data your analysts actually ask about (ratings, findings, risk vectors, vendor intelligence and threat intelligence), encrypt traffic end to end, keep your prompts out of model training, and let you move from read-only questions to controlled write actions when you are ready.

Must-have features

  • Hosted remote MCP endpoint with OAuth sign-in so analysts can connect from Claude, Cursor, VS Code or ChatGPT without running their own server.
  • Role-respecting identity model that inherits each user's existing permissions, access groups and product subscriptions rather than granting a shared service account.
  • Broad data coverage across security ratings, findings, risk vectors, vendor portfolios, vulnerability context and threat intelligence, exposed through the vendor's full API surface.
  • Encrypted traffic and a secure proxy design between the assistant, the MCP server and the underlying platform.
  • A written commitment that API calls, prompts and responses are not used to train AI or machine learning models.
  • Support for both read and write operations with explicit, auditable controls over which actions the agent can take.

Bitsight's MCP server meets each of these criteria out of the box. It is hosted at https://mcp.bitsighttech.com/mcp, supports OAuth through the assistant's connector settings, and is listed in the Claude Connector Directory so Claude users can add it in a few clicks. It uses your own Bitsight identity and honors your access controls, access groups and product subscriptions, so it never grants broader access than you already have. The seven core tools let the assistant discover and call any of Bitsight's 1,000+ API operations covering ratings, findings, risk vectors, vendor intelligence, vulnerability context and threat intelligence. It runs as a secure proxy with encrypted traffic, we do not use MCP traffic to train AI or machine learning models, and it supports both read and write operations so you can scale from Q&A to agentic workflows on your own terms.

How to set up the Bitsight MCP server in Claude, ChatGPT, and Cursor

Before you connect anything, confirm two prerequisites. First, you need an active Bitsight subscription to at least one product the MCP can surface data from: Security Posture Management (SPM), Continuous Monitoring (CM), Vendor Risk Management (VRM), or Bitsight Cyber Threat Intelligence (CTI). The MCP uses your own Bitsight identity and respects your existing role-based access controls, access groups, and product subscriptions, so it will not grant anything you do not already have. Second, you need an MCP-compatible assistant: Claude, ChatGPT, or an IDE such as Cursor, VS Code, or Kiro. In every client below, the server URL is the same: https://mcp.bitsighttech.com/mcp.

Setting up Bitsight MCP in Claude (directory and Claude Code)

  1. Open the Claude Connector Directory at claude.ai/directory/bitsight.
  2. Click Connect (or Add) and review the permissions Claude will request.
  3. Approve the connector, then sign in to Bitsight when the OAuth window opens. Claude never sees your Bitsight password.
  4. Return to Claude, open Customize > Connectors, and confirm that bitsight-mcp is listed and enabled.
  5. For Claude Code, create a project-scoped .mcp.json file in your project folder, or run: claude mcp add --scope project bitsight-mcp --transport http https://mcp.bitsighttech.com/mcp --header "X-Bitsight-API-Token: <api-token>". Use --scope user to make it available across all projects.

The directory listing is the fastest path for most analysts because it handles the OAuth sign-in for you, and you choose whether to connect Bitsight, CTI, or both. Claude Code is the better option for engineers who want the connection scoped to a specific repository and checked into version control with teammates. If you are on a free Claude account you can only add one custom connector, and enterprise IT may disable custom connectors entirely, so coordinate with your admin before you start.

Setting up Bitsight MCP in Cursor, VS Code, and Kiro

  1. Open the MCP configuration file for your IDE: ~/.cursor/mcp.json for Cursor, ~/.vscode/mcp.json for VS Code, or ~/.kiro/settings/mcp.json for Kiro.
  2. Add an entry named bitsight-mcp with the transport set to http and the URL https://mcp.bitsighttech.com/mcp.
  3. Save the file. Without any headers, the IDE will open a browser window for OAuth sign-in the first time a tool is called.
  4. Sign in to Bitsight and, when prompted, choose Bitsight, CTI, or both.
  5. Reopen the MCP panel in the IDE and confirm that bitsight-mcp shows a green connected state with its tools enabled.

OAuth is the simplest path in an IDE because the browser handles the sign-in and token refresh for you. If your workflow needs headless access, for example to run the MCP inside a long-running coding agent, the same config file accepts an X-Bitsight-API-Token header for SPM, CM, and VRM, plus X-CTI-Client-ID and X-CTI-Client-Secret for CTI. We recommend header-based credentials in any environment where the IDE cannot open a browser window reliably.

Setting up Bitsight MCP in ChatGPT Developer mode

  1. Confirm you are on a ChatGPT Business, Enterprise, or Edu workspace. Developer mode and custom MCP apps are not available on Free, Plus, or Pro plans, and your workspace admin controls whether they are enabled.
  2. Open Settings > Apps > Advanced Settings and turn on Developer mode.
  3. Choose Create app, name it Bitsight MCP, and enter https://mcp.bitsighttech.com/mcp as the server URL.
  4. Select an authentication method: OAuth, or Access token / API key using the custom header X-Bitsight-API-Token.
  5. Click Create (paste your API token on the next screen if you chose a token), then confirm the MCP shows as connected.

OAuth is the easiest option for individual analysts and keeps the connection tied to each user's Bitsight identity. If you use the access-token option, keep tokens one-to-one with named users so audit logs stay meaningful.

Verify the connection. In any client, open a fresh chat and ask the assistant to list the Bitsight tools it can call. You should see the seven core MCP tools, covering API discovery and schemas, Knowledge Base search, and read and write operations against more than 1,000 Bitsight API endpoints. Follow up with a low-risk read query such as "Show me my portfolio's current ratings" to confirm the OAuth session is active and traffic is flowing through the encrypted proxy to Bitsight.

How third-party risk and GRC teams monitor vendors with Bitsight MCP

Third-party risk, governance risk and compliance (GRC), and security operations teams use the Bitsight MCP server to turn vendor monitoring from a weekly review ritual into something closer to a continuous conversation. The outcomes we hear most often: shorter time from signal to action, fewer manual pivots between the Bitsight platform and spreadsheets, and executive-ready narratives that explain why a vendor rating moved without an analyst rewriting them by hand. The six approaches below are what these teams run against their portfolios today.

  • Ask conversational questions of your full vendor portfolio: the Bitsight MCP server and its invoke_query_api tool read ratings, findings and alerts, so an analyst can ask 'List the vendors in my portfolio with a rating drop this month' and get an answer without writing a query.
  • Run a preconfigured supply chain readiness assessment: the Mythos Supply-Chain Readiness workflow in Bitsight AI Mode helps teams assess vendor exposure across the portfolio in one agentic pass.
  • Measure program performance for the board: the Analyze TPRM ROI and Impact workflow in AI Mode reports vendors monitored, portfolio mean rating, EVAs sent, risk outcomes and recommended actions.
  • Codify repeat tasks as reusable Bitsight Skills: a SKILL.md file (optionally with scripts or templates) installed from the in-platform Skills Library or the public Bitsight skills repository on GitHub, so the next analyst runs the same declining-vendors or resilience review the same way.
  • Hunt for vendor exposure on underground forums: the Threat Hunting Intel Items API, called through Bitsight MCP, lets an analyst query conversationally, inspect returned posts, diagnose false-positive patterns and iterate complex searches in a single session.
  • Bring your own security posture into the same assistant: Bitsight Security Posture Management data flows into ChatGPT, Claude or Cursor through Bitsight MCP, then gets wrapped in Bitsight Skills so posture-trend analysis becomes a repeatable AI workflow alongside vendor monitoring.

What ties these approaches together is the data underneath the assistant. Bitsight combines third-party risk management with exposure management, continuous monitoring and cyber threat intelligence, all powered by Bitsight AI. That matters for MCP-driven vendor monitoring because the model is only as useful as the dataset it can reach. When an agent can see your vendor ratings, your external exposure, continuous monitoring signals and underground chatter from one authenticated connection, the answers stop being partial. You can ask a single question about a single vendor and get a single, defensible picture back.

Best practices and expert tips for MCP-driven vendor risk monitoring

Our approach to MCP-driven vendor risk monitoring is simple: give AI agents fast, authenticated access to Bitsight data, keep humans in charge of decisions that touch vendors, and treat every third-party MCP server or skill as untrusted until it is proven otherwise. The tips below reflect how we designed the Bitsight MCP and what our TRACE threat research team has seen in the wild in 2026.

  • Keep a human in the loop before any vendor outreach: Let the agent gather ratings, findings and alerts, draft the summary and even the email, but route the escalation to a named analyst for review. AI is a force multiplier for triage, not a substitute for the judgment call that starts a vendor conversation.
  • Require explicit confirmation for every write operation: The Bitsight MCP blocks write calls on the first attempt and runs them only after a single-use confirmation. Leave that default on, and reserve the 24-hour waiver for narrowly scoped bulk work by trusted operators.
  • Audit and constrain MCP extensions and AI skills before install: Bitsight TRACE identified roughly 7,600 fraudulent GitHub repositories distributing SmartLoader malware, with more than 800 disguised as AI skills or MCP servers (including a lookalike Docker MCP Gateway) to get indexed in public AI registries. Pin extensions to known-good sources such as github.com/Bitsight/skills, review SKILL.md contents and scripts, and block side-loaded servers from developer machines.
  • Prefer authenticated, vendor-hosted MCP servers: TRACE identified nearly 1,000 internet-accessible MCP servers with no authorization in place, acting as open gateways between AI agents and back-end systems. Use the Bitsight-hosted endpoint over OAuth or API headers, and never expose an unauthenticated MCP server to the internet.
  • Turn recurring queries into reusable Bitsight Skills: When a prompt proves useful more than twice, package it as a Skill with a SKILL.md file and optional scripts, and install it in your agent's skills directory. That gives every analyst the same repeatable answer and removes prompt drift between team members.
  • Use AI Mode administrator controls to govern what agents can do: You can opt out of Chats, Agentic Workflows and Skills independently in Account Preferences, and each Skill's SKILL.md states whether it runs through the Bitsight MCP or directly against the Bitsight APIs. Start narrow, expand by use case, and revisit the settings each quarter as your program matures.

Advantages and benefits of using an MCP server for vendor risk monitoring

Once the Bitsight MCP is wired into your assistant, the payoff shows up in places third-party risk and GRC teams can actually measure: hours saved on API plumbing, faster triage of alert queues, and board-ready narratives that no longer wait on a human analyst to assemble them. The five benefits below are the ones we hear about most often from teams that run Bitsight MCP in Claude, ChatGPT, Cursor and Claude Code.

  • Plain-language access to 1,000+ Bitsight API operations without writing code: The Bitsight MCP gives an MCP-compatible assistant direct access to your Bitsight data and the full Bitsight API, so analysts can ask questions in natural language instead of hand-coding calls. A single connection can discover and invoke any of 1,000+ operations across ratings, findings, risk vectors, vendor intelligence, vulnerability context and threat intelligence, using the permissions the user already has.
  • Machine-speed consumption of cyber risk intelligence for agentic GRC workflows: Because Bitsight supports MCP access patterns, AI agents and agentic GRC workflows can pull vendor risk data continuously, without a human pausing to log in, export a CSV and share it. That lets scheduled checks, triage bots and ticket-routing agents work at machine speed on the same data your analysts see.
  • Permission-respecting identity model that avoids granting broader access than the user already has: The MCP exposes seven core tools for discovery, schema inspection, Knowledge Base search and read and write calls, and it binds every one of them to the customer's existing Bitsight identity. It respects role-based access controls, access groups and product subscriptions, so connecting an assistant never widens what a given user can see or do.
  • Privacy guarantees built into the transport layer: We do not use API calls, prompts or responses passing through the MCP to train any AI or machine learning model. The MCP runs as a secure proxy with all traffic between the MCP and Bitsight encrypted, which keeps vendor names, findings and portfolio detail out of model training sets and off unprotected channels.
  • Faster alert triage and repeatable executive-ready summaries on demand: By deploying MCPs and Skills in the AI environment of your choice, agentic workflows help teams move from reviewing intelligence to automatically applying it, including querying Bitsight's full dataset in natural language and drafting executive-ready summaries and rating-change narratives on demand. Alert queues that used to need a dedicated analyst can be pre-filtered, scored and summarized before a human opens the ticket.

How Bitsight simplifies MCP-based vendor risk monitoring

We are the only vendor that brings third-party risk management together with exposure management, continuous monitoring and cyber threat intelligence, all powered by Bitsight AI. That matters for MCP-based vendor risk monitoring because your assistant pulls from one connected intelligence layer instead of stitched-together point tools. Bitsight AI delivers three capabilities that work together: Bitsight MCP connects compatible external AI assistants and agents to your Bitsight data and APIs, AI Mode is the native in-platform experience with chats and agentic workflows, and Bitsight Skills package reusable instructions your agent can follow consistently across vendors.

These pieces are built to fit how risk and security teams actually work. Bitsight MCP supports both read and write operations, so Claude, Cursor or ChatGPT can look up ratings, findings and alerts and, with your confirmation, update the portfolio. Bitsight Skills work either through Bitsight MCP or directly against Bitsight APIs, which means the same workflow runs in an IDE or on a schedule. AI Mode lets you opt out of Chats, Agentic Workflows and Skills, and Bitsight Vendor Risk Management anchors the full lifecycle from onboarding through continuous monitoring and offboarding. If you want to see this running against your own portfolio, request a Bitsight demo.

The future of vendor risk monitoring with MCP and agentic AI

MCP adoption is reshaping how third-party risk and security teams consume intelligence. Since the protocol was introduced in November 2024, vendor risk has become one of its most useful applications. The risk surface is widening at the same pace: Bitsight TRACE researchers identified roughly 1,000 internet-accessible MCP servers running with no authorization, and were able to enumerate the tools those servers exposed. That is why we built Bitsight MCP, AI Mode and Bitsight Skills as a vendor-backed path for continuous monitoring that keeps your identity, controls and data boundaries intact. Talk to the Bitsight team to see the MCP server in action.