mcp-audit
by P4ST4S
README.md
# mcp-audit
[](https://github.com/P4ST4S/mcp-audit/actions/workflows/ci.yml)
[](https://codecov.io/gh/P4ST4S/mcp-audit)
[](https://glama.ai/mcp/servers/P4ST4S/mcp-audit)
[](https://github.com/P4ST4S/mcp-audit/discussions)
A drop-in security and observability proxy for MCP servers. `mcp-audit` sits between an MCP client and any upstream MCP server to produce signed audit trails, redact sensitive payloads, enforce allow/deny policies and per-tool rate limits, and expose a local read-only dashboard.
For a contributor-oriented map of the runtime, package boundaries, concurrency model, and design invariants, see [ARCHITECTURE.md](ARCHITECTURE.md).
## Why mcp-audit?
[The MCP 2026 roadmap](https://modelcontextprotocol.io/development/roadmap) calls out enterprise needs around audit trails, gateway patterns, and operational visibility. `mcp-audit` fills that gap as a deployable sidecar or local wrapper: it sits between any MCP client and server, preserves protocol traffic, and records signed audit entries for tool calls, resource reads, prompt requests, and all other JSON-RPC methods.
```text
+-------------+ JSON-RPC / MCP +-----------+ JSON-RPC / MCP +-------------+
| MCP client | <-------------------> | mcp-audit | <---------------------> | MCP server |
+-------------+ +-----------+ +-------------+
|
v
JSONL or SQLite audit log
|
v
Read-only dashboard
```
## What This Is / Is Not
`mcp-audit` is not a domain-specific MCP server. It is a transparent security and observability proxy that wraps any MCP server and audits the JSON-RPC traffic passing through it.
Directories may show the tools exposed by the upstream server, not tools implemented by `mcp-audit` itself.
## Supported Transports
- `stdio` for local MCP clients such as Claude Desktop
- `http` for MCP servers exposed over HTTP
HTTP upstreams can use custom CA bundles, TLS server name overrides, and optional mTLS client certificates. Upstream retries are disabled by default and only apply to conservative, idempotent JSON-RPC methods when enabled; `tools/call` is not retried.
## Use Cases
- Audit tool calls made by AI agents in regulated environments
- Detect unexpected or dangerous MCP tool usage
- Keep signed JSONL or SQLite logs for incident review
- Redact sensitive fields before storing requests and responses
- Block disallowed tools and apply per-tool rate limits without modifying the upstream MCP server
## Demo
## Install
For detailed platform-specific instructions and troubleshooting, see
[INSTALL.md](INSTALL.md).
Download the latest prebuilt binary and run:
```bash
# Linux/macOS: resolve latest, download, verify it starts
version=$(curl -fsSL https://api.github.com/repos/P4ST4S/mcp-audit/releases/latest \
| grep '"tag_name"' | head -n1 | cut -d'"' -f4 | sed 's/^v//')
os=$(uname | tr '[:upper:]' '[:lower:]')
arch=$(uname -m); [ "$arch" = "x86_64" ] && arch=amd64 || arch=arm64
base="https://github.com/P4ST4S/mcp-audit/releases/download/v${version}"
archive="mcp-audit_${version}_${os}_${arch}.tar.gz"
curl -L -o "${archive}" "${base}/${archive}"
tar -xzf "${archive}"
./mcp-audit --version
```
Run with Docker:
```bash
docker run --rm ghcr.io/p4st4s/mcp-audit:latest --version
```
Install from source with Go:
```bash
go install github.com/P4ST4S/mcp-audit/cmd/mcp-audit@latest
```
To pin a specific release for reproducible installs, see
[INSTALL.md](INSTALL.md#choosing-a-version).
## Quick Start
Run in stdio mode:
```bash
AUDIT_SECRET="$(openssl rand -hex 32)" \
mcp-audit --transport stdio --upstream "npx @modelcontextprotocol/server-filesystem /tmp"
```
On Windows PowerShell, generate the secret and set it as an environment variable:
```powershell
$env:AUDIT_SECRET = -join ((1..32) | ForEach-Object { '{0:x2}' -f (Get-Random -Max 256) })
.\mcp-audit.exe --transport stdio --upstream "npx @modelcontextprotocol/server-filesystem C:\Temp"
```
Run in HTTP mode:
```bash
mcp-audit --transport http --upstream http://localhost:8080 --port 4422
```
Run with Docker Compose:
```bash
docker compose up --build
```
The dashboard is available at `http://127.0.0.1:9090` by default.
Prometheus metrics are available at `http://localhost:9091/metrics` by default.
## Examples
- [Cursor stdio configuration](examples/cursor/README.md)
- [Continue stdio configuration](examples/continue/README.md)
- [VS Code stdio configuration](examples/vscode/README.md)
- [Claude Desktop stdio configuration](examples/claude-desktop/README.md)
- [Zed stdio configuration](examples/zed/README.md)
## Configuration
`mcp-audit` loads `config.yaml` from the current directory by default. CLI flags override config values, and `AUDIT_SECRET` overrides `audit.secret`.
| Key | Default | Description |
| --- | --- | --- |
| `proxy.transport` | `stdio` | Proxy transport: `stdio` or `http`. |
| `proxy.upstream` | required | Stdio command or HTTP upstream URL. |
| `proxy.port` | `4422` | HTTP listen port. |
| `proxy.upstream_timeout_ms` | `30000` | HTTP upstream request timeout in milliseconds. |
| `proxy.forward_headers` | empty | Request headers allowed to bypass the default upstream strip list. Use `["Authorization"]` only when the upstream MCP HTTP server requires bearer-token auth. |
| `proxy.tls.ca_file` | empty | Optional CA bundle used to verify an HTTPS upstream MCP server. |
| `proxy.tls.server_name` | empty | Optional TLS server name override for the upstream MCP server. |
| `proxy.tls.insecure_skip_verify` | `false` | Skip upstream TLS certificate verification. Intended only for local testing. |
| `proxy.tls.client_cert_file` | empty | Optional client certificate for upstream mTLS. Must be configured with `proxy.tls.client_key_file`. |
| `proxy.tls.client_key_file` | empty | Optional client key for upstream mTLS. Must be configured with `proxy.tls.client_cert_file`. |
| `proxy.retry.max_retries` | `0` | Maximum conservative retry attempts for safe HTTP upstream requests. Off by default. |
| `proxy.retry.initial_interval_ms` | `200` | Initial upstream retry backoff. |
| `proxy.retry.max_interval_ms` | `2000` | Maximum upstream retry backoff. |
| `proxy.client_id` | `claude-desktop` | Client identifier written to audit entries. |
| `proxy.server_id` | `filesystem` | Server identifier written to audit entries. |
| `audit.storage` | `jsonl` | Storage backend: `jsonl` or `sqlite`. |
| `audit.path` | `./audit.jsonl` | JSONL audit log path. |
| `audit.sqlite_path` | `./audit.db` | SQLite database path. |
| `audit.sign` | `true` | Enable HMAC-SHA256 signatures when a secret is set. |
| `audit.secret` | empty | HMAC secret. Prefer `AUDIT_SECRET`. |
| `audit.async.enabled` | `false` | Enable asynchronous batched audit writes through a bounded ring buffer. |
| `audit.async.queue_size` | `4096` | Maximum queued audit entries before backpressure blocks writers. |
| `audit.async.batch_size` | `128` | Maximum entries written per storage batch. |
| `audit.async.flush_interval_ms` | `1000` | Maximum time before a partial batch is flushed. |
| `audit.rotation.max_size_bytes` | `0` | Maximum active JSONL file size before archive rotation. `0` disables built-in rotation. |
| `audit.rotation.max_files` | `0` | Maximum number of rotated JSONL archives to keep. `0` disables retention. |
| `audit.rotation.interval` | empty | Optional time-based JSONL rotation interval: `hourly` or `daily`. Empty disables time-based rotation. |
| `audit.rotation.max_age_days` | `0` | Delete JSONL archives whose filename rotation timestamp is older than this many days. `0` disables age retention. |
| `middleware.rate_limit.enabled` | `true` | Enable per-client, per-tool token buckets. |
| `middleware.rate_limit.requests_per_minute` | `60` | Allowed requests per minute per `(client_id, tool_name)`. |
| `middleware.redact.enabled` | `true` | Enable JSON key-based PII redaction. |
| `middleware.redact.patterns` | sensitive keys | Case-insensitive key fragments to redact. |
| `policy.enabled` | `false` | Enable synchronous allow/deny policy checks for `tools/call`. |
| `policy.default_action` | `allow` | Fallback action when no policy rule matches: `allow` or `deny`. |
| `policy.rules` | empty | Ordered first-match allow/deny rules for tool calls. |
| `dashboard.enabled` | `true` | Serve the dashboard. |
| `dashboard.bind_address` | `127.0.0.1` | Dashboard listen address. Set explicitly, for example to `0.0.0.0`, only when the dashboard is protected by network controls or auth. |
| `dashboard.port` | `9090` | Dashboard listen port. |
| `dashboard.auth.token` | empty | Optional bearer token required as `Authorization: Bearer <token>` for dashboard HTML and API requests. |
| `metrics.enabled` | `true` | Serve Prometheus metrics on a separate HTTP endpoint. |
| `metrics.port` | `9091` | Metrics listen port. |
| `metrics.path` | `/metrics` | Metrics HTTP path. |
| `metrics.include_go_metrics` | `true` | Include Go runtime metrics. |
| `metrics.include_process_metrics` | `true` | Include process metrics. |
| `metrics.tool_labels` | `true` | Include `tool_name` and `client_id` labels for tool-level metrics. Disable to minimize label cardinality. |
| `otel.enabled` | `false` | Export `tools/call` audit entries as OTLP/HTTP JSON spans. |
| `otel.endpoint` | `http://localhost:4318` | OTLP HTTP endpoint base URL. `/v1/traces` is appended automatically. |
| `otel.service_name` | `mcp-audit` | OpenTelemetry `service.name` resource attribute. |
| `otel.headers` | empty | Additional OTLP HTTP headers, for example `Authorization` or API key headers. |
| `otel.tls.ca_file` | empty | Optional CA bundle used to verify the OTLP endpoint. |
| `otel.tls.server_name` | empty | Optional TLS server name override. |
| `otel.tls.insecure_skip_verify` | `false` | Skip OTLP TLS certificate verification. Intended only for local testing. |
| `otel.retry.max_retries` | `3` | Maximum OTLP retry attempts after a failed export request. |
| `otel.retry.initial_interval_ms` | `200` | Initial OTLP retry backoff. |
| `otel.retry.max_interval_ms` | `2000` | Maximum OTLP retry backoff. |
| `otel.queue_size` | `1024` | Maximum queued audit entries before trace exports are dropped. |
| `otel.batch_size` | `64` | Maximum spans per OTLP export request. |
| `otel.flush_interval_ms` | `1000` | Maximum time before a partial OTLP batch is exported. |
| `otel.timeout_ms` | `5000` | OTLP HTTP request timeout. |
By default, `mcp-audit` strips hop-by-hop request headers and `Authorization` before forwarding HTTP requests to the upstream. To pass a bearer token to a trusted authenticated upstream, opt in explicitly:
```yaml
proxy:
forward_headers:
- Authorization
```
Security note: forwarded headers, including secrets like bearer tokens, are transmitted verbatim to the upstream server. Only enable this if you control or trust the upstream MCP server. `Authorization` is the only sensitive header that can be opt-in forwarded because some MCP HTTP servers require it for upstream authentication. `Cookie`, `Set-Cookie`, and `Proxy-Authorization` are always rejected: they represent state destined for other components such as browser sessions or proxy chains and have no legitimate use in MCP request forwarding. If an existing deployment relied on implicit `Authorization` forwarding, add the config above.
JSONL rotation is disabled by default and supports size-based and UTC time-based triggers. Rotated archives use UTC timestamps such as `audit.jsonl.20260610T214605Z`; if multiple rotations happen in the same second, numeric suffixes are added. The archive timestamp reflects the wall-clock time of the rotation event, not the cutoff that was crossed. Time-based rotation is append-driven: `mcp-audit` does not start a background timer, so if no writes occur for several days, the active file is not rotated until the next append after the cutoff. Missed cutoffs are not caught up; the next append creates at most one archive.
```yaml
audit:
storage: jsonl
rotation:
max_size_bytes: 104857600
interval: daily
max_files: 10
max_age_days: 30
```
`max_age_days` uses the rotation timestamp encoded in the archive filename. This means age since rotation, not the age of the oldest entry inside the archive. Age retention runs before `max_files` retention. Compression and SQLite archival are not part of this release.
CLI flags:
```text
--transport stdio | http
--upstream upstream server command or URL
--port proxy port for http mode
--upstream-timeout upstream HTTP request timeout in milliseconds
--config path to config.yaml
--storage jsonl | sqlite
--no-dashboard disable the web dashboard
--no-metrics disable Prometheus metrics
--version print version and exit
--log-level debug | info | warn | error
```
## Claude Desktop
Configure Claude Desktop to spawn `mcp-audit` instead of the upstream MCP server:
```json
{
"mcpServers": {
"filesystem-audited": {
"command": "mcp-audit",
"args": [
"--transport",
"stdio",
"--upstream",
"npx @modelcontextprotocol/server-filesystem /tmp"
],
"env": {
"AUDIT_SECRET": "replace-with-a-long-random-secret"
}
}
}
}
```
## Dashboard
The dashboard shows recent entries, filters, expandable request/result JSON, top tools, calls today, and error rate. It refreshes every five seconds.
By default the dashboard listens only on `127.0.0.1:9090`. To expose it on another interface, configure `dashboard.bind_address` explicitly and enable authentication or place it behind a trusted access proxy.
```yaml
dashboard:
enabled: true
bind_address: 127.0.0.1
port: 9090
auth:
token: "replace-with-a-long-random-token"
```
When `dashboard.auth.token` is configured, requests to `/`, `/api/entries`, and `/api/stats` must include:
```text
Authorization: Bearer replace-with-a-long-random-token
```
Missing or invalid credentials return `401 Unauthorized` with `WWW-Authenticate: Bearer realm="mcp-audit-dashboard"`. Repeated failed authentication attempts from the same remote address are throttled with `429 Too Many Requests`.
Dashboard JSON API responses include `Cache-Control: no-store` so intermediaries and browsers do not retain audit payloads.
## Prometheus Metrics
`mcp-audit` exposes Prometheus metrics on a separate endpoint so platform teams can scrape operational data without exposing the dashboard.
```yaml
scrape_configs:
- job_name: mcp-audit
static_configs:
- targets: ["localhost:9091"]
```
Application metrics use the `mcp_audit_` prefix and avoid unbounded labels. Tool-level labels can be disabled with `metrics.tool_labels: false` for stricter cardinality control. Policy decisions are exposed as `mcp_audit_policy_decisions_total{action="allow|deny"}`.
For a ready-made Prometheus + Grafana stack, see [examples/docker-compose-observability](examples/docker-compose-observability/README.md).
## Policy Engine
`mcp-audit` can enforce synchronous allow/deny rules before a `tools/call` reaches the upstream MCP server. Denied calls return a JSON-RPC error and are still written to the audit log.
```yaml
policy:
enabled: true
default_action: allow
rules:
- action: deny
client_id: claude-desktop
server_id: filesystem
tool_name: delete_file
reason: "Destructive filesystem operations are blocked"
```
Rules are evaluated in order. Empty fields and `*` match any value, so `default_action: deny` can be used with explicit allow rules for stricter deployments.
## OpenTelemetry
`mcp-audit` can export `tools/call` audit entries as OTLP/HTTP JSON spans to Jaeger, Tempo, Honeycomb, or any OTLP-compatible collector.
```yaml
otel:
enabled: true
endpoint: "http://localhost:4318"
service_name: "mcp-audit"
headers:
Authorization: "Bearer your-token"
timeout_ms: 5000
retry:
max_retries: 3
initial_interval_ms: 200
max_interval_ms: 2000
```
The exporter uses current OpenTelemetry MCP and GenAI semantic conventions where possible, including `mcp.method.name`, `jsonrpc.request.id`, `gen_ai.operation.name`, `gen_ai.tool.name`, `network.transport`, `network.protocol.name`, `rpc.response.status_code`, and `error.type`. Project-specific attributes are kept link-oriented, such as `mcp_audit.entry_id`, `mcp_audit.direction`, `mcp_audit.client_id`, `mcp_audit.server_id`, `mcp_audit.storage`, and `mcp_audit.signature.present`.
Request params and tool results are not exported to spans by default. The signed JSONL or SQLite audit row remains the evidence artifact; OTLP provides correlation, latency, and operational visibility.
Exporter health is visible through Prometheus metrics under the `mcp_audit_otel_` prefix, including export requests, span outcomes, dropped spans, queue depth, and queue capacity. Temporary OTLP failures are retried with bounded exponential backoff; `Retry-After` is honored for retryable responses up to `otel.retry.max_interval_ms`.
## Audit Entries
Each stored entry includes a ULID, timestamp, direction, transport, JSON-RPC method, tool name when present, redacted params/result, JSON-RPC error when present, duration, client/server identifiers, and an optional HMAC-SHA256 signature.
Example JSONL entry:
```json
{
"id": "01HY8G6Y8S6W9K6ZD7VJ4Q8X4R",
"timestamp": "2026-05-25T12:34:56Z",
"direction": "client_to_server",
"transport": "stdio",
"method": "tools/call",
"tool_name": "read_file",
"params": {
"name": "read_file",
"arguments": {
"path": "/tmp/example.txt",
"token": "[REDACTED]"
}
},
"duration_ms": 18,
"client_id": "claude-desktop",
"server_id": "filesystem",
"signature": "hmac-sha256:..."
}
```
The signature covers:
```text
id + timestamp + method + tool_name + raw_params
```
## Roadmap
- SIEM-friendly exports
- OTLP compression and trace context propagation
## Contributing
Keep changes small, run `go build ./...` and `go vet ./...`, and prefer standard library behavior over new dependencies. Stability guarantees are documented in [STABILITY.md](STABILITY.md).
See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, PR expectations, and project principles. See [CHANGELOG.md](CHANGELOG.md) for release history.
## Community
- [Discussions](https://github.com/P4ST4S/mcp-audit/discussions): questions, ideas, and design conversations
- [Issues](https://github.com/P4ST4S/mcp-audit/issues): bug reports and concrete feature requests
- Security: see [SECURITY.md](SECURITY.md) for the private vulnerability reporting process
## License
Apache-2.0. See [LICENSE](LICENSE).
TDQS
A3.6/5.0
Scored across 14 tools
Disambiguation2/5
read_file is a deprecated duplicate of read_text_file, and list_directory/list_directory_with_sizes/directory_tree all provide overlapping directory listings. Most other tools are distinct, but these boundary cases create real selection ambiguity for agents.
Naming Consistency4/5
Tool names overwhelmingly follow a snake_case verb_noun pattern like read_, write_, edit_, list_, move_, search_, and get_. Minor deviations are directory_tree lacking a verb and the deprecated read_file alongside read_text_file, but overall the naming is predictable.
Tool Count4/5
14 tools is within a reasonable scope for a filesystem server, but the set is slightly padded by the deprecated read_file and the near-duplicate list_directory_with_sizes. Removing or merging those would make the count feel tighter without losing functionality.
Completeness2/5
The server covers reading, writing, editing, moving, searching, and listing, but has no delete_file, delete_directory, or copy operation. Since write and edit are present, the lack of delete and copy leaves significant file-management dead ends for agents.
Maintenance
ActivityInactive
ResponsivenessResponsive