# GRAFOMEM CGR Capture [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/GNS-Foundation/grafomem  
**GitHub Stars:** 3  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/grafomem-cgr-capture

## Description
Record agent judgments and their real outcomes on GRAFOMEM Cloud. Capture now, score later.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "grafomem-cgr-capture": {
    "command": "uvx",
    "args": ["grafomem"]
  }
}
```

## Documentation

## What GRAFOMEM CGR Capture MCP server does

The GRAFOMEM CGR Capture MCP server connects agent workflows with GRAFOMEM Cloud to record judgments and the outcomes those judgments eventually produce. Its purpose is to preserve the relationship between a decision made by an agent and the real-world result used to evaluate that decision.

This supports GRAFOMEM’s Capability-Grounded Reputation model. CGR evaluates reputation from judgments that resolve against outcomes, rather than treating reputation as an unsupported claim. The captured records provide the evidence base for later scoring and review.

The service is distinct from the broader GRAFOMEM runtime described in the repository. That runtime also covers governed memory, signed checkpoints, content-addressed state, and cryptographic erasure. The capture listing focuses on judgment and outcome records hosted on GRAFOMEM Cloud.

## How it works

A workflow submits an agent judgment for capture, then records the relevant real outcome when it becomes available. The system can therefore separate the time of the prediction from the time of resolution. This makes it suitable for workflows where a decision cannot be assessed immediately.

The resulting evidence can feed CGR scoring. The repository describes CGR as a model in which reputation is earned per domain from judgments that later resolve, with peer reviews weighted according to the reviewer’s demonstrated calibration. The associated cgr-bench project is described as an independently reproducible implementation of the scoring model and its evaluation properties.

GRAFOMEM also describes governed memory as an evidence substrate: facts can carry provenance, and memory operations can be governed and verified. Those broader properties belong to the GRAFOMEM system; the supplied material does not define which of them are exposed specifically by this capture server.

## Tools and capabilities

The documented capabilities are:

- Capture agent judgments in GRAFOMEM Cloud.
- Record the real outcomes associated with earlier judgments.
- Preserve decision evidence for later CGR scoring.
- Support reputation evaluation based on resolved judgments rather than assertions.
- Provide an MCP integration for agent clients described in the repository as Claude and MCP.

The provided excerpt does not list formal MCP tool names, argument schemas, return values, authentication fields, or outcome-resolution states. Developers should consult the linked integration documentation before implementing against a specific interface.

## Limitations and notes

The material does not state whether this capture service is self-hosted, delivered only through GRAFOMEM Cloud, or packaged separately from the reference REST and MCP server. It also does not document environment variables, installation steps, transport details, authentication requirements, retention behavior, or supported MCP clients beyond the repository’s general Claude/MCP reference.

The repository mentions a hosted GRAFOMEM Cloud free tier of 10,000 governed decisions per month, but the excerpt does not explicitly say that this quota applies to every use of the GRAFOMEM CGR Capture MCP server. It also does not describe automatic scoring at capture time; the stated workflow is to capture judgments first and score them later.

_Full upstream README: https://allmcps.com/mcp/grafomem-cgr-capture/readme_