Evidence · bounded receipt proof

How to Record MCP Tool Calls | Reelier

Evidence state: traced against released reelier@0.32.0 on 2026-08-12.

▸ Direct answer

Place Reelier's recorder between the agent and a stdio MCP server. Calls that traverse that proxy are recorded. Supported command configuration can be rewritten, but plugin-owned manifests and URL entries are not rewritten, so those calls remain outside unless the plugin author wraps its stdio entry or the host exposes supported configuration.

Evidence note

The pinned initialization, wrapper, and recorder implementations trace how supported command entries are wrapped and how traversing MCP calls are recorded.

Proxy placement

The wrapper fronts a stdio server command. The agent talks to the proxy and the proxy talks to the server, giving the recorder visibility into calls that traverse that process boundary.

Supported configuration

Installation can rewrite supported command-based MCP configuration and retain a backup. URL entries are skipped because the wrapper itself speaks stdio.

Calls the trace does not cover

Plugin-owned manifests load outside the supported host configuration unless the plugin author declares a wrapped stdio entry. Direct HTTP, native shell actions, and file edits outside the proxy are not captured by this path.

Non-claims

This does not prove that an action was safe or semantically correct.

This does not prove that every write was observed or receipted.

Falsifier

A call that traverses the configured stdio proxy but is absent from the resulting trace would falsify this claim.

Citation-friendly definition

Reelier records MCP calls that traverse its stdio proxy; traffic outside that configured proxy is outside the trace's observation boundary.

Sources

  1. Pinned configuration discovery
  2. Pinned stdio wrapper
  3. Pinned recorder implementation

Reading as an agent? This page has a markdown twin: /evidence/record-mcp-tool-calls/md