S
supermax01/fireboat
Give AI agents safe, read-only access to one patient's FHIR health records over MCP.
updated 2026-11-18
View on GitHub ↗Link checked 2026-11-17 · HTTP 200 · opens the vendor's own site
- Star count
- 0 ★ · supermax01/fireboat · read from the GitHub search API · (checked 2026-11-19 20:00:48)
- Source
- read directly from its GitHub page · type: open-source project
- Harvest history
- first seen 2026-11-16 · last seen 2026-11-20 · seen 5× · listed by 2 sources
- Score
- score 203 · 2 sources · repo
FAQ (answers come from our records)
Who is it for?
We filed it under these tags: Open Source. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository
supermax01/fireboat;first catalogued 2026-11-16;last seen 2026-11-20。There is no official certification or quality rating here — we record sources and dates only.README excerpt (English, untranslated) · Read the full README on GitHub →
# fireboat
Excerpted from the repository README (abridged)
Give AI agents safe, read-only access to one patient's FHIR health records over MCP.
> Early preview, not a release. Uses synthetic (made-up) patient data only.
> Runs locally on your own machine. Read-only. No SMART on FHIR sign-in yet.
> Not clinically validated. Not affiliated with or endorsed by HL7.
> No npm package or release yet.
Contents: What it is · Who it is for · Why · How it works ·
Try it · Use it with an AI agent · The five tools ·
What a response looks like · When something is not allowed ·
Configuration · Project layout · Development and testing ·
What works today · Questions
## What it is
fireboat sits between an AI agent, such as Claude, and a FHIR server, the standard format for electronic health records.
It lets the agent search and read one patient's records through MCP (Model Context Protocol), the standard way AI agents connect to tools.
The patient is chosen by whoever sets fireboat up, never by the AI.
Before anything reaches the AI, fireboat checks that it belongs to that patient and removes what the AI should not see.
Today it focuses on one job: looking up a patient's recent lab results.
## Who it is for
Developers building AI agents on FHIR data who want a worked example of letting an agent read health records without
letting it wander: a fixed patient, no writes, and answers that say what they could and could not check.
People evaluating how AI should handle clinical data, who want to see where the safety checks sit and run them
against synthetic data.
Anyone curious about MCP who wants a small, heavily tested server to read and learn from.
It is not for production use or real patients. See What works today.
## Why
One patient, locked in: the patient is set in configuration, so the AI cannot ask for anyone else. Every record that comes back is checked again.
Read-only: five tools for finding and reading records. Nothing can write, change or delete data.
Exact values: a lab value written as 7.0 stays 7.0. A number that cannot be kept exactly is refused, never rounded.
Honest answers: every result says what was checked and whether the list may be incomplete.
Size limits: every response has a maximum size, so a large record cannot flood the AI's context.
Heavily tested: thousands of automated tests, and the tests are themselves checked by deliberately breaking the code to make sure they notice (a full run of those checks has not passed yet).
The security model and testing guide have the details.
## How it works
mermaid
flowchart TD
agent["AI agent Claude Code or another MCP host"]
config[/"Your configuration which patient, which server"/]
subgraph fb["fireboat, running on your machine"]
tools["Five read-only tools"]
checks["Request checks patient comes from configuration only allowed search filters no URLs from the agent"]
http["HTTP client only the configured local server no redirects, size and time limits"]
recheck["Every returned record is checked against the patient again"]
clean["Hidden data removed patient identifiers, attachment contents"]
shape["Answer shaped exact numbers, coverage labels"]
limit["Response size limit"]
store[("Short-lived result store for paging")]
end
fhir[("FHIR R4 server synthetic data")]
> Early preview, not a release. Uses synthetic (made-up) patient data only.
> Runs locally on your own machine. Read-only. No SMART on FHIR sign-in yet.
> Not clinically validated. Not affiliated with or endorsed by HL7.
> No npm package or release yet.
Contents: What it is · Who it is for · Why · How it works ·
Try it · Use it with an AI agent · The five tools ·
What a response looks like · When something is not allowed ·
Configuration · Project layout · Development and testing ·
What works today · Questions
## What it is
fireboat sits between an AI agent, such as Claude, and a FHIR server, the standard format for electronic health records.
It lets the agent search and read one patient's records through MCP (Model Context Protocol), the standard way AI agents connect to tools.
The patient is chosen by whoever sets fireboat up, never by the AI.
Before anything reaches the AI, fireboat checks that it belongs to that patient and removes what the AI should not see.
Today it focuses on one job: looking up a patient's recent lab results.
## Who it is for
Developers building AI agents on FHIR data who want a worked example of letting an agent read health records without
letting it wander: a fixed patient, no writes, and answers that say what they could and could not check.
People evaluating how AI should handle clinical data, who want to see where the safety checks sit and run them
against synthetic data.
Anyone curious about MCP who wants a small, heavily tested server to read and learn from.
It is not for production use or real patients. See What works today.
## Why
One patient, locked in: the patient is set in configuration, so the AI cannot ask for anyone else. Every record that comes back is checked again.
Read-only: five tools for finding and reading records. Nothing can write, change or delete data.
Exact values: a lab value written as 7.0 stays 7.0. A number that cannot be kept exactly is refused, never rounded.
Honest answers: every result says what was checked and whether the list may be incomplete.
Size limits: every response has a maximum size, so a large record cannot flood the AI's context.
Heavily tested: thousands of automated tests, and the tests are themselves checked by deliberately breaking the code to make sure they notice (a full run of those checks has not passed yet).
The security model and testing guide have the details.
## How it works
mermaid
flowchart TD
agent["AI agent Claude Code or another MCP host"]
config[/"Your configuration which patient, which server"/]
subgraph fb["fireboat, running on your machine"]
tools["Five read-only tools"]
checks["Request checks patient comes from configuration only allowed search filters no URLs from the agent"]
http["HTTP client only the configured local server no redirects, size and time limits"]
recheck["Every returned record is checked against the patient again"]
clean["Hidden data removed patient identifiers, attachment contents"]
shape["Answer shaped exact numbers, coverage labels"]
limit["Response size limit"]
store[("Short-lived result store for paging")]
end
fhir[("FHIR R4 server synthetic data")]
Listed badge (optional)
Copy this line into the README if you want the listing mark. Optional, and it does not affect being listed.
[](https://anclum.com/en/s/supermax01-fireboat/)
- Source
- github · https://github.com/supermax01/fireboat
- Browse by field
- Open source
- Listed by
- 2 independent sources
- added
- today
More entries
Filed under
Every tag opens the full list for that filter. Counts are the live catalogue figures.