Skip to main content
The Trassets MCP server lets AI clients read your Data API v1 data through the Model Context Protocol. Every call runs with your API key and is subject to the same restrictions, grants, and rate limit as a direct API request.

​
What the MCP server can do

The server offers five tools. All of them only read data; each one calls a Data API endpoint with your key.

​
Endpoint and key

You need a key with channel MCP. An administrator creates it in the portal like any other key (see Authentication) and selects MCP under Channel. A key works for exactly one channel: the MCP server rejects keys created for the API, and the REST API rejects MCP keys.
Data that a tool returns is processed by the model provider of your AI client. Only connect clients whose provider your organization has approved for this data, and don’t combine the MCP server with tools that can send data to third parties. See Risks.

​
Supported AI clients

MCP is model-independent: the client brings the model, so the model that answers depends on your client. What matters is whether the client can send a fixed API key header to a remote MCP server. The MCP server does not support OAuth sign-in yet, so clients that only offer OAuth can’t connect. As of November 2026. Vendors change their MCP support often; if a step doesn’t match, check the client’s documentation.

​
Example: Claude Code

Then ask Claude Code which Trassets tables it can read. Other clients need the same two values: the URL and the key, sent as X-API-Key header or as Authorization: Bearer <key>. Don’t put the plaintext key into a config file you commit: Cursor resolves ${env:TRASSETS_MCP_KEY} in headers, and VS Code prompts for it through ${input:...}.

​
Example: Microsoft Copilot Studio

  1. Open your agent, go to Tools, select Add a tool, and add a new Model Context Protocol server.
  2. Enter a name, a description, and https://mcp.trassets.ai/mcp as server URL.
  3. Under authentication, choose API key, type Header, and enter X-API-Key as header name.
  4. Create a connection, enter your MCP key, and select Add to agent. All five tools are enabled by default.
  5. Before you publish the agent, check which credentials the tool uses. With maker-provided credentials, everyone who uses the agent reads with your key and sees everything your key may see. Instead, have an administrator create one MCP key per user and let each user connect with their own key.
Then ask the agent, for example, “Which Trassets tables can you read?” or “How many properties are in the data?”

​
Limits

  • Read only. The server can’t create or change data.
  • No calculations. Tools return rows, not sums, averages, or KPIs. If you ask for a total, the model in your client computes it from the rows it has read, and it can miscount or stop after one page. For row counts, the tools return total. Pre-computed figures are only those in the KPI tables.
  • Small pages. A tool call returns at most 200 rows (default 50). Reading a large table takes many calls, and each call counts against your key’s budget of 2,500 requests per hour (see Rate limits). For full exports, use the REST API with a separate key for channel API (see Pagination).
  • Exact filters only. get_table_data filters by exact value on a table’s filter columns. Ranges and text search are not available.
  • Data freshness. The data is as current as the last sync of each table; there is no fixed interval. get_changes shows last_sync_at per table (see Sync cycle).
  • API key only. There is no OAuth sign-in yet (see Supported AI clients).

​
Risks

  • Data leaves Trassets. Everything a tool returns goes to the model provider of your client and is processed under the provider’s terms. Your organization decides which provider may process which data. Use keys without the personal data grants unless the use case needs them.
  • Instructions hidden in data. Free-text fields such as comments or booking texts can contain text that reads like an instruction. The tools tell the model to ignore such text, but no model follows that reliably.
  • Outbound tools. An agent that reads your data, processes such text, and can also send data out (email, web requests, file uploads, MCP servers that write elsewhere) can be tricked into sending your data to third parties. Use the Trassets MCP server only in agents and sessions without such tools.
  • Unverified answers. The model can misread fields, mix up tables, or extrapolate from one page. Check figures against the API or your reports before you rely on them.
  • One key per person. Create one MCP key per person, name it after them in the portal, and don’t share it, not even through an agent that others use. When someone leaves, delete their key; that ends their MCP access.

​
Error messages

Tools report errors as text. The most common ones: