Assurance
Checks what an agent did not read, with coverage over a retrieval step or folder
- Pricing
- Open source
- Platforms
- CLI, macOS, Windows
- Runs
- Locally
- Version
- 0.5.10
- Listing updated
- today
Similar MCP servers
About Assurance
Assurance is an open-source MCP server for developer tools. It runs locally on your machine and is started by your AI client.
Assurance starts from the gap between an agent saying it is done and what it actually checked. This MCP server answers a narrow version of that question: given a retrieval step or a folder, what was never read. The parent project also audits Claude Code and Codex sessions after the fact, reporting repeated failures, edits with no recognised test run afterwards, and commands it cannot classify.
The README is careful about claims. It checks only what the last prompt names, says whether a named test ran after the last edit, and states plainly that it cannot tell whether the work is correct. A companion command reads git history to name files that usually change together with the ones you touched.
The tool has no server, account or telemetry of its own. The MCP package assurance-mcp installs from PyPI and runs over stdio. The use-case pages cover payments, banking, healthcare, legal and platform work.
Install
Run in a terminal
claude mcp add assurance-mcp -- uvx assurance-mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"assurance-mcp": {
"command": "uvx",
"args": [
"assurance-mcp"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"assurance-mcp": {
"command": "uvx",
"args": [
"assurance-mcp"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"assurance-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"assurance-mcp"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"assurance-mcp": {
"command": "uvx",
"args": [
"assurance-mcp"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"assurance-mcp": {
"command": "uvx",
"args": [
"assurance-mcp"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Unread coverage
- Reports what was not read in a retrieval step or folder.
- Session audit
- Summarises what a Claude Code or Codex session did and skipped.
- Test-after-edit check
- Flags edits with no recognised test run afterwards.
- Co-change hints
- Uses git history to name files that usually change together.
- Local only
- No server, account or telemetry.
- Hook option
- Can run after every Claude Code turn.
Good for
- Check whether a research agent read every document in a folder
- Find missing reporting periods in a set of files
- Audit what a coding agent skipped before merging
- Spot a forgotten file in a change
Pricing
Open source
Assurance FAQ
Does it send data anywhere?
The README says the tool has no server, account or telemetry of its own.
Which agents are audited?
Claude Code and Codex sessions, plus your own agent's run record or traces.
Does it say the work is correct?
No, it says it does not claim to know whether the fix is right.
Is Assurance free?
Yes. Assurance is open source, so you can use and self-host it for free.
How do I add Assurance to Claude Code?
Run this command in your terminal: claude mcp add assurance-mcp -- uvx assurance-mcp
What are the best alternatives to Assurance?
The closest alternatives to Assurance are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


