
ResiliReplay
io.github.aliengineering-bytePublished by GitHub account aliengineering-bytev0.7.1Updated Oct 24, 2026
Run bounded, deterministic MCP reliability tests and generate executable regressions.
Overview
ResiliReplay is a local MCP server that runs bounded, deterministic reliability tests against MCP servers and turns failures into executable regression tests.
- What you can do
- Run a bundled demo that reproduces a deterministic failure, retries once, and records duplicate effects without any config or credentials. Test a reviewed MCP server from a config file with a dry-run plan, then execute it after approval by plan hash. Tune runs with JSON output, output paths, regression toggling, timeouts and retry limits. Generate and execute regression tests from causal failures and print an evidence digest. Keep the existing audit workflow and persist evidence as metadata and hashes.
- When to use it
- Use it when you want to check how an MCP server behaves under injected failures, verify bounded recovery and absence of duplicate effects, or turn observed failures into repeatable regression tests before relying on that server.
- Requirements
- Runs locally over stdio via npx from the npm package resilireplay. Node.js 22 or 24 is supported. Testing a real server needs a config file and server name; no account, credential, paid model or remote target is required for the demo.
Installation
In SourceWeft
- Open ResiliReplay in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
ResiliReplay
Inject deterministic MCP failures, verify bounded recovery and duplicate-effect behavior, and turn causal failures into executable regression tests.
The demo is a bundled local fixture: no config, account, credential, paid model, telemetry, or remote target. It runs a clean call, reproduces one deterministic failure, retries once, records zero duplicate effects, generates and executes a regression, prints an evidence digest, and cleans up.
Test a reviewed server without starting it:
Repeat the displayed plan with --approve <plan-sha256> to execute it. --json, --output,
--no-regression, --timeout, and --retries map to the same bounded engine. The existing
mcp audit workflow remains supported.
Node.js 22 and 24 are supported. Evidence persists metadata and hashes instead of unrestricted tool bodies or credentials. Remote targets retain explicit ownership controls.
Agent-runtime capture, campaigns, replay, Studio, adapters, MCP-RES v0.1/v0.2, security guidance, and the real packed-package MCP Everything example are documented in the repository README.
Apache-2.0.
Source: packages/cli/README.md at commit f101725
Tools
0Version history
1- v0.7.1LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts