Platform MCP server

The platform MCP server lets your agent work on your Kapa projects as you. Once connected, Claude Code, Codex, Cursor, ChatGPT, or Claude can add knowledge sources, check what they ingested, run test queries, create integrations, and pull analytics, all with your own Kapa permissions.

Connect your agent

Paste this prompt into your agent. The agent installs the Kapa plugin if it is not connected yet, asks you to sign in, and then tells you what your project holds and what it can do next.

ChatGPT and Claude cannot install the plugin themselves. Connect Kapa first, as described in If your agent needs help connecting, then paste the prompt.

Agent prompt

Then ask for what you want done, for example "add our documentation site to Kapa". For a guided first setup, follow Set up Kapa with your agent.

Connection

Every Kapa user connects to the same URL:

https://mcp.kapa.ai/mcp

Your client needs to support remote MCP servers and a browser sign-in. If your agent needs help connecting has the configuration for each client.

The Kapa plugin adds the kapa-setup-* skills on top, one per source type, so the agent knows the call sequence for each source rather than only the tool descriptions.

Authentication

The server accepts Kapa accounts only. On the first connection, your client opens a consent page in your browser, then the Kapa login. There is no separate Google or GitHub sign-in.

Tools act as you. list_projects returns your project permissions on each project, and a call that needs a permission you lack on that project is rejected. No tool takes a Kapa API key or other Kapa credential. Tools that connect a source take that source's own credentials, such as a Confluence token, which the agent asks you for. When a call fails to authenticate, sign in to the server again from your client; in Claude Code, that is /mcp.

Tools

Every tool except the three server tools calls a Kapa platform endpoint as you. A tool takes only the inputs an agent should set, which can be fewer than the endpoint accepts, and the tool's description in the server's tool list is the reference for its parameters. search_project_knowledge, list_integrations, and the analytics tools match their endpoints in the API reference. The tool list comes from Kapa and can change without an update to your client, so the table groups the tools by area rather than listing every one.

Start with list_projects: it returns every project in your team, with your permissions on each one. Tools that act on an existing project or object take its ID. Validation and discovery tools take only the connection details of a source.

Area	Tools
Projects	list_projects
Sources	list_sources, get_source, update_source, duplicate_source, refresh_source, delete_source, and list_source_groups
Source setup	create_<type>_source for each source type the agent can set up; get_<type>_config, set_<type>_config, and update_<type>_config to read and write a source's configuration; and connect_<type> and configure_<type> for sources you authorize in the browser (Google Drive, Salesforce, and Zendesk)
Source checks	validate_<type> tools that check access to the third-party system, check_<type>_connection for Google Drive and Zendesk, and list_<type>_* tools that list what a connector can read, such as spaces, labels, categories, or folders
Web crawling	preview_crawl, list_preview_pages, start_crawl, start_crawl_with_review, get_crawl_status, and cancel_crawl; get_crawl_config, set_crawl_config, and update_crawl_config; the content selector tools detect_content_selector, inspect_content_selector, get_content_selector, set_content_selector, and update_content_selector; and the review tools list_pages_for_review and approve_pages_for_review
Custom Answers	list_custom_qa_pairs, add_custom_qa_pair, update_custom_qa_pair, and delete_custom_qa_pair
Retrieval	search_project_knowledge, the same agentic retrieval that the hosted MCP server and the HTTP API expose
Integrations and API keys	list_integrations; a create_<type>_integration tool for the Website Widget, Slack Bot, Discord Bot, Support Form Deflector, hosted MCP server, API, and Agent SDK integrations; and create_api_key and list_api_keys
Analytics	get_project_activity, list_threads, get_thread, list_end_users, list_top_questions_periods, get_top_questions, list_coverage_gaps_periods, and get_coverage_gaps

Searches made with search_project_knowledge are recorded in Conversations like any other query. Kapa creates an integration for them in the project the first time the server searches it, named Kapa Platform MCP by default. It does not count toward your plan's integration limit and does not complete the Deploy an integration step under Get started.

Server tools

Three tools belong to the server itself rather than to a platform endpoint:

• search_kapa_docs searches Kapa's own documentation, so the agent can look up how a feature works before configuring it. It never searches the sources of your projects.
• give_context sends Kapa what you are trying to accomplish and why. The server's instructions tell the agent to call it once at the start of a task, and to leave out names, proprietary information, personal data, and credentials.
• give_feedback reports a problem with the tools to Kapa: an error, a missing capability, confusing output, or unclear documentation. It is separate from the hosted MCP server's feedback tool, which records feedback for your own team.

Both give_context and give_feedback record your Kapa user, email address, and team with what the agent sends.