Build an integration

TKP stores uploaded documents, images, audio and video as resources. Resources belong to libraries, which group content and control who can access it. Each library belongs to a tenant, the workspace shared by a team. Your application can search the libraries it can read and retrieve the sources behind each result. See tenants and library access for how these boundaries work.

Before connecting, get the TKP application URL and access to a library. This documentation site may use a different hostname. Ask the operator for the application URL if you only have this site.

Choose your interface

Use case	Start here
Make HTTP requests from your application to upload, search and retrieve files	REST quickstart
Give an agent tools to search TKP and inspect source content	MCP quickstart
Connect an existing application that supports MCP	Client configurations
Ask a coding agent to build your integration	Agent context and downloads

REST endpoints start at /api/v1 on the application URL. The MCP endpoint is /mcp on the same application. Both use the same libraries and access permissions.

Grant access

For a script or backend service, sign in to TKP and open your workspace's Access Tokens page. Create a token, enable API for REST or MCP for MCP, and grant Read for one library. Also grant Create resources if the integration will upload files. Save the token and copy its plaintext value before leaving the page; it is shown only once. Store it in your application's environment or secret store.

A token belongs to one tenant and cannot exceed its owner's library permissions. A valid token can therefore return no libraries if it has no effective read access. See permissions and abilities before granting broader access.

If users should connect their own TKP accounts, use OAuth instead of asking them to copy tokens. OAuth opens a browser consent screen where the user chooses the workspace and permissions for the connection.

Follow the document lifecycle

1. Select a library the token or OAuth connection can read.
2. Upload a file if you need to add content.
3. Wait for processing to finish before searching the uploaded resource.
4. Search and inspect the relevant source excerpts or original files.
5. Keep source identifiers and links with any answer your application displays.

TKP calls file processing ingestion. It extracts and indexes content and may use the instance's configured AI providers. An accepted upload is not yet a searchable resource. The quickstarts and workflow guides explain which processing states to wait for and when to stop on a failure.

Inspect the contract

The application's /api/documentation page describes REST endpoints interactively. Its /docs?api-docs.json URL serves an OpenAPI schema, a machine-readable description of those endpoints and their fields. The reference downloads describe the version shipped with this documentation site.

For MCP, a client calls tools/list after connecting to discover tool names and input schemas, which describe the accepted arguments. That list reflects the current credential's permissions and can be smaller than the full tool catalog.