Connect Anthropic to Claude Desktop
Call Claude models, count tokens before you spend them, and read organization usage and cost. Add Toolspoke to Claude Desktop once and it can use 8 Anthropic actions. Your Anthropic credential stays in Toolspoke, and you choose which actions Claude Desktop may call.
Set it up in Claude Desktop
Three steps, the first one once for every toolkit you will ever add.
Add Toolspoke to Claude Desktop
Settings, then Connectors, then Add custom connector
https://toolspoke.com/mcpPaste the address, then approve the request when Toolspoke asks.
Install Anthropic and connect it
Add Anthropic to a project in your Toolspoke workspace.
API key. You provide it once, when you install the connector. Toolspoke encrypts it at rest and decrypts it only for the length of a single call, and the gateway attaches it to the outbound request itself, so it is never part of the arguments an agent sends.
- API key
- platform.claude.com → Settings → API keys
Choose what Claude Desktop may do
All 8 actions start on. Switch off the one action that changes something in Anthropic, and Claude Desktop gets a refusal if it tries.
What you can ask Claude Desktop for
A few of the Anthropic actions that only read, so they are safe to try first.
- List the organization's workspaces with their ids, names, display colours and archive state
- What the organization was actually billed, in USD, bucketed by day
What Claude Desktop can do in Anthropic
7 actions read and one writes.
Reads
List models
List the Claude models this API key can call, newest first, with each model's id, display name, release date, context window and maximum output tokens. Call this before create_message or count_tokens whenever a model id is not certain, this connector deliberately does not carry a hard-coded list of model names, because Anthropic's roster changes faster than a file in a repository can. Free, and the cheapest proof that the key works.
Get model
Read one model by its id: display name, release date, context window, maximum output tokens, and the capability flags saying whether it accepts images or PDFs, supports extended thinking, structured outputs, citations, code execution or the Batch API. Use it to check that a model supports a feature before building a request around that feature, and to confirm an id from list_models is still available to this key. A 404 means the id does not exist for this key, call list_models to see what does.
Count tokens
Count how many input tokens a message request would use, without sending it and without being billed for the completion. Takes the same `model`, `messages`, `system` and `tools` that create_message takes, so the honest way to use it is to build the request, count it here, and only then decide whether to spend it. A read, because nothing is created and nothing is charged. Returns a single input_tokens number; it does not predict output tokens, which are the larger part of the cost of most calls.
List workspaces
List the organization's workspaces with their ids, names, display colours and archive state. Needs the separate Admin API key, not the ordinary one, without an admin_key credential this call is refused before it is sent. The `wrkspc_…` ids it returns are what usage_report and cost_report take to break a report down by workspace. The Admin API is unavailable to individual accounts; on one of those this returns nothing useful no matter which key is pasted.
List api keys
List the organization's API keys as records, id, name, status, partial key hint, which workspace each belongs to and who created it. It does not and cannot return key values; Anthropic shows those once, at creation. Needs the separate Admin API key. The `apikey_…` ids are what usage_report takes to attribute token spend to a particular key, which is the usual reason to call this.
Usage report
Token usage for the organization over a time range, in buckets, split into uncached input, cache reads, cache writes and output tokens. This is the accurate number, far better than adding up what individual responses reported. Group by "model", "workspace_id", "api_key_id" or "service_tier" to see where the tokens went; filter with models, workspace_ids or api_key_ids to narrow it. Needs the separate Admin API key. Bucket limits are Anthropic's: 1d allows at most 31 buckets per call, 1h at most 168, 1m at most 1440, ask for a range that fits, or page with `page` when has_more comes back true. For money rather than tokens, call cost_report.
Cost report
What the organization was actually billed, in USD, bucketed by day. Prefer usage_report when the question is about tokens and this when the question is about money. Group by "workspace_id" to attribute spend to a team, or "description" to see it split by line item, grouping by description is also how web search and code execution charges become visible, since they are not token usage. Needs the separate Admin API key. Daily buckets only; Anthropic rejects any other bucket width here. Priority Tier spend is billed differently and does not appear at all, track that through usage_report instead.
Writes
Create message
Generate a response from a Claude model. This is the operation that costs money: it is billed per input and output token, which is why it is a write while count_tokens and list_models are reads, and why it is not this connector's health check. Streaming is out of scope here, the response comes back whole, with its content blocks, stop reason and token usage. `max_tokens` is required and caps the output; set it to what the answer actually needs rather than to the model's ceiling. Call list_models for a valid `model` id and count_tokens to price the request first. The response is a model's own output and the prompt is prose, so both are kept out of the audit log.
Anthropic in Claude Desktop: questions
Does Claude Desktop see my Anthropic credential?
No. The credential is stored encrypted in Toolspoke, and the gateway adds it to each call. Claude Desktop only ever talks to Toolspoke, and the dashboard shows the last four characters of the credential and nothing more.
Can I stop Claude Desktop from changing things in Anthropic?
Yes. Each of the 8 actions can be switched off for a project. Turn off the one that changes something and Claude Desktop can only read.
Will 8 actions crowd Claude Desktop's context?
No. Toolspoke shows Claude Desktop five bridge tools, plus a guide where a toolkit has one, and Claude Desktop searches for the Anthropic action it needs when it needs it. Adding more toolkits does not add more tools to its list.
Does it work in Claude on the web too?
Yes. A connector added in Settings belongs to your Claude account, so it appears in Claude on the web and in the desktop app.
What does it cost?
Tool calls are free on every plan, and a Anthropic call does not run in a sandbox, so it bills nothing.
Anthropic in your other apps
More ai toolkits for Claude Desktop
Try Anthropic in Claude Desktop
Free to start, and tool calls cost nothing on any plan.