For agents and developers

Agent interfaces

ToolTwist publishes its own content in machine-readable form. An AI agent can query our case studies, services, and FAQs, and register interest in a call, without a human parsing HTML. This page documents those interfaces: the endpoints, what each tool does, the limits, and what happens to anything you send us.

Everything described here is live. If something is missing, wrong, or you need a capability we do not expose, email sales@tooltwist.com.

Three ways in

MCP server
/api/mcp
Connect an MCP client over streamable HTTP. Tools, resources, and prompts, no authentication.
WebMCP, in the browser
document.modelContext
Every page registers the same tools for a page-local agent. Calls proxy back to the MCP server, so there is one implementation, not two.
Text corpus
/llms.txt · /api/md
The whole site as Markdown for retrieval, plus a per-page Markdown twin for any route that has one.

Connecting

Endpoint: https://tooltwist.com/api/mcp
Transport: Streamable HTTP (JSON-RPC 2.0 over POST)
Authentication: None. The surface is read-mostly and public.
Sessions: Not required — tools/list and tools/call work statelessly.
Server card: https://tooltwist.com/.well-known/mcp.json
Discovery: Every response carries Link headers for the server card, the MCP endpoint, and llms.txt; the same relations appear as <link> elements in the page head.

A GET to the endpoint returns 405 — this server does not offer a server-initiated event stream, which the MCP specification permits. Send a POST.

Tool catalogue (6 tools)

All but one are read-only. Each tool carries MCP behaviour annotations, so a client can decide what needs a human in the loop — the table below states the same values the server sends.

Tool	Arguments	Effect
`list_case_studies` List all ToolTwist portfolio case studies with slug, title, and summary.	no arguments	Read-only, repeatable
`get_case_study` Get the full content of a ToolTwist case study as Markdown, by slug.	`slug` — Case study slug, e.g. property-hub	Read-only, repeatable
`get_services` List ToolTwist's service and industry pages with one-line summaries.	no arguments	Read-only, repeatable
`get_contact_info` Get ToolTwist contact details: emails, phone, and offices.	no arguments	Read-only, repeatable
`search_content` Full-text search across ToolTwist case studies, FAQs, and service pages. Returns matching pages with a snippet.	`query` — Search terms	Read-only, repeatable
`start_discovery_call` Capture intent to book a discovery call with ToolTwist. The team follows up by email.	`name` — Requester's name `email` — Requester's email address `company`(optional) — Company name (optional) `message` — What they want to build or discuss	Writes — submits a lead. Not repeatable: two calls create two enquiries.

Prompts

evaluate_tooltwist
Assess whether ToolTwist's services and delivery record fit a described project.
- project_description — What the user wants to build, in their own words
summarize_case_studies_for_industry
Compare the ToolTwist case studies relevant to a given industry.
- industry — Industry or sector, e.g. banking

Resources

site://overview
Who ToolTwist is, where the offices are, and what we do.
site://faq
The full FAQ as Markdown.
site://contact
Emails, phone numbers, and office addresses.

Limits, errors, and your data

Rate limits

The read-only tools are not rate limited. start_discovery_call is capped at 5 calls per hour per source address. Over the cap, the call returns a normal tool result with isError set and text naming the retry window — not an HTTP error, so an agent can read and act on it.

Validation

Submitted fields are validated before anything else happens: name up to 120 characters, a syntactically valid email address, optional company up to 160, and a message between 1 and 5000 characters. A rejected submission comes back as a tool result explaining which field to fix.

What happens to a submitted enquiry

There is no database behind this site. A valid start_discovery_call is formatted as an email and sent to our sales inbox through Amazon SES, with the address you supplied set as the reply-to. It follows exactly the same path as an enquiry from the website — the agent route is not treated differently.

If our mailer is unavailable, the tool tells you so in its reply and gives you the direct address instead, so a request is never silently dropped.

Only send personal details you are authorised to share, and tell the person you are acting for that you are sharing them. Handling is covered by our Privacy Policy; to have an enquiry deleted, email info@tooltwist.com.

Crawling

AI crawlers are welcome — our robots.txt names them explicitly and allows the whole site. Prefer llms.txt or the MCP server over scraping HTML: both are cheaper for you and stay in sync with the pages.