Content
# Fluid MCP
[![CI][ci-badge]][ci-workflow]
[ci-badge]: https://github.com/botnetbuddies/fluid-mcp/actions/workflows/ci.yml/badge.svg
[ci-workflow]: https://github.com/botnetbuddies/fluid-mcp/actions/workflows/ci.yml
Fluid MCP is a standalone MCP stdio server that renders operator-reviewed
commands from the [`botnetbuddies/cheats`](https://github.com/botnetbuddies/cheats)
with CheatMD ([website](https://cheatmd.dev),
[repository](https://github.com/cheatmd-dev/cheatmd)).
It keeps a cached cheats checkout and uses imported CheatMD Go packages to parse,
index, lint, and render commands in-process. Render/search surfaces remain
non-executing.
## First-use flow
1. Install or run Fluid MCP with either the local Go binary or Docker.
2. Add the matching command to your MCP client configuration.
3. Start the client and use Fluid MCP tools to search, inspect, render, review,
and scope-check commands.
4. Keep render-only output under human operator review.
## Install or run
Requirements for source installs:
- Go 1.26.3+
- `git` on `PATH`
### Local binary
Install the latest released source build:
```bash
go install github.com/botnetbuddies/fluid-mcp/cmd/fluid-mcp@latest
```
Install the current pinned release:
```bash
go install github.com/botnetbuddies/fluid-mcp/cmd/fluid-mcp@v0.2.1
```
Prebuilt release archives and SHA-256 checksums are published at
[GitHub Releases](https://github.com/botnetbuddies/fluid-mcp/releases/tag/v0.2.1).
Run the installed server over stdio:
```bash
fluid-mcp
```
Run from a local checkout:
```bash
go run ./cmd/fluid-mcp
```
### Docker
Build the image:
```bash
docker build -t fluid-mcp .
```
Run with a persistent cheats cache:
```bash
docker run -i --rm -v fluid-mcp-cache:/cache fluid-mcp
```
Smoke-test startup without an MCP client:
```bash
docker run -i --rm -v fluid-mcp-cache:/cache fluid-mcp < /dev/null
```
The smoke test exits when stdin closes.
## Configure your MCP client
Fluid MCP speaks MCP over stdio. Pick the configuration that matches how you run
it. Fluid MCP writes diagnostics to stderr so stdout stays reserved for MCP
protocol traffic.
### Local binary configuration
Use this when `fluid-mcp` is installed on your `PATH`:
```json
{
"command": "fluid-mcp",
"args": []
}
```
### Docker configuration
Use this when the MCP client should launch the Docker container:
```json
{
"command": "docker",
"args": ["run", "-i", "--rm", "-v", "fluid-mcp-cache:/cache", "fluid-mcp"]
}
```
First startup needs network access to clone the cheats repository. Later starts
reuse the cache and update it when GitHub is reachable. See the
[configuration guide](docs/configuration.md) for flags, environment variables,
cache behavior, and troubleshooting.
## First safe use
Typical render-only flow:
1. Search for relevant cheats, methods, or chains.
2. Inspect the selected entry and its source/provenance metadata.
3. Provide required variables and optional engagement scope.
4. Review the rendered command or recipe.
5. Run anything outside Fluid MCP only when it is authorized and appropriate.
Scope-aware rendering can block command generation before command text is
returned. Pass `engagement_scope` plus `scope_context` to generation tools, or
call `evaluate_engagement_scope` directly before rendering. Scope checks are
guardrails for operator review; they are not a substitute for reading the signed
ROE.
For the full safe workflow, scope model, methods/chains, and evidence reporting,
read the [operator workflow guide](docs/operator-workflow.md). For the complete
MCP tool list and response metadata, read the [tool reference](docs/tool-reference.md).
## Documentation
- [Configuration and troubleshooting](docs/configuration.md)
- [Safe operator workflow](docs/operator-workflow.md)
- [MCP tool reference](docs/tool-reference.md)
- [Verification and releases](docs/release.md)
- [Manual MCP smoke tests](PILOT_TESTING.md)
- [Changelog](CHANGELOG.md)
## Contributing, security, license
- Contributions: [CONTRIBUTING.md](CONTRIBUTING.md)
- Security reporting and operational guidance: [SECURITY.md](SECURITY.md)
- License: MIT, see [LICENSE](LICENSE)
Connection Info
You Might Also Like
Train-in-Silence
The first Task-Aware MCP server and automated VRAM calculator for LLM...
stacklit
108,000 lines of code. 4,000 tokens of index. One command makes any repo...
AppClaw
AI-powered mobile automation agent — describe what you want in plain...
pdf-mcp
Production-ready MCP server for PDF processing with intelligent caching....
kotadb
Local-only code intelligence API for AI developer workflows (Bun +...
gemini-api-docs-mcp
A remote HTTP MCP server for searching Google Gemini API documentation.