Vascue Public Knowledge Search README
The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Vascue Public Knowledge Search listing page.
Vascue Public Knowledge Search (MCP server)
A Model Context Protocol server that searches Vascue's public documentation: healthcare operations, the AI front desk for clinics, provider-side insurance claims automation, Cliniko integration, security, case studies and pricing.
It comes in two equivalent forms:
-
Hosted (always current):
https://www.vascue.io/mcp/search- streamable HTTP, no authentication. -
Self-contained (this repo):
python server.py- local BM25 search over a bundled snapshot of the public pages (content/, refreshed per release withscripts/fetch_content.py). No network calls at runtime, so it also works offline and is what directory-built releases run. -
Endpoint:
https://www.vascue.io/mcp/search(streamable HTTP, no authentication) -
Server card: https://www.vascue.io/.well-known/mcp/server-card.json
-
Registry name:
io.vascue/public-knowledge-search -
Operated by: Vascue Limited (ISO 27001 certified)
Public content only. This server indexes public product and educational pages. Never send patient information, claim documents, clinic credentials or booking requests to it. Agent-based clinic booking is a separate research pilot, not a public API.
Connect
Any MCP client that speaks streamable HTTP can connect to the endpoint directly.
Claude Code
Cursor / Claude Desktop / other stdio-only clients (via mcp-remote)
Self-contained local server (stdio; bundled snapshot, no network)
Docker (builds the self-contained server)
Tools
One tool, no authentication, read-only.
search
Hybrid (keyword + vector) search over Vascue's public pages. Returns matching excerpts with their canonical https://www.vascue.io/... URLs so answers can cite the source.
| Input | Type | Notes |
|---|---|---|
query | string (required) | Natural-language question or keywords, e.g. "how does Vascue handle insurance claim pre-authorisation". |
ai_search_options.retrieval.retrieval_type | "hybrid" | "vector" | "keyword" | Default hybrid. |
ai_search_options.retrieval.max_num_results | integer 1–50 | Default 8. |
ai_search_options.retrieval.match_threshold | number 0–1 | Default 0.35. |
ai_search_options.retrieval.context_expansion | integer 0–3 | Neighbouring chunks to include. |
Query rewriting and reranking are disabled server-side; the server returns source chunks only and never a generated answer, so nothing is presented as a Vascue statement without a citation. Rate limit: 60 requests per minute per client.
Example call:
The endpoint is backed by a Cloudflare AI Search instance over the approved public Markdown export of vascue.io (the service descriptor at https://www.vascue.io/.well-known/ai-search.json states what is and is not indexed).
Development
CI runs the same build and smoke test on every push and weekly, so the badge above doubles as an endpoint health indicator.
Directory build specs
Directories that build the server from source (e.g. Glama) run the self-contained form. Generated build images vary (uv-managed Python without pip, or a PEP 668 externally-managed system Python), so use an explicit venv:
- Build steps:
["uv venv /opt/venv && uv pip install --python /opt/venv/bin/python -r requirements.txt"] - CMD:
["/opt/venv/bin/python", "server.py"] - No environment variables.
Where a normal pip exists, plain pip install -r requirements.txt + ["python", "server.py"] works too.
Other machine-readable surfaces
https://www.vascue.io/llms.txthttps://www.vascue.io/openapi.json(public, read-only content API)https://www.vascue.io/.well-known/agent-skills/index.json(agent skills; also at vascue-io/skills)
Licence
This repository (README, manifest, Dockerfile) is MIT licensed. The content served by the endpoint is Vascue's public website content.