Skip to main content

dcert-mcp

CI/CD Pipeline dcert-mcp on PyPI dcert on PyPI Python versions License: MIT

dcert-mcp is a Model Context Protocol server that gives AI assistants the dcert certificate and TLS tools. The server is a prebuilt Rust executable (dcert-mcp), installed as a wheel script, so starting it runs no Python. The optional proxy extra adds a FastMCP proxy, typed async client and the dcert-mcp-proxy command. Tools added to the Rust server are discovered at runtime, so the Python layer never needs to change.

This package is separate from dcert, the command line tool. Installing dcert-mcp pulls in dcert at the same version and nothing else.

Installation

Command Result Third party Python packages
pip install dcert-mcp the dcert-mcp MCP server, and dcert through the dependency 0
pip install "dcert-mcp[proxy]" plus the FastMCP proxy, typed client and dcert-mcp-proxy the FastMCP stack
pip install "dcert-mcp[proxy,otel]" plus OpenTelemetry export the FastMCP stack and OpenTelemetry

Requires Python 3.12 or later. Wheels exist for Linux x86_64 (glibc 2.34 or later: RHEL 9, Ubuntu 22.04, Debian 12 and later), macOS Intel, macOS Apple Silicon and Windows x86_64. Only these platform wheels are published, so an install from PyPI, a private mirror or a downloaded wheel never fetches anything from GitHub. On any other platform pip reports "no matching distribution"; build the binaries from source and point DCERT_MCP_BINARY at the server.

For a reproducible install of the extra, requirements/proxy.txt (and otel.txt for [proxy,otel]) in the repository pins every package with hashes for all platforms: pip install --require-hashes -r proxy.txt next to pip install dcert-mcp.

Without the extra, the Rust server works on its own (dcert-mcp speaks MCP over stdio, or HTTP with --mode http). The Python helpers below need the extra: importing one without it fails with a single line naming it, for example dcert_mcp needs the optional package 'fastmcp'. Install it with: pip install "dcert-mcp[proxy]".

Quick start

As a server

from dcert_mcp import create_server

create_server().run()  # stdio transport

As a client

import asyncio
from dcert_mcp import create_client

async def main() -> None:
    async with create_client() as client:
        tools = await client.list_tools()
        result = await client.call_tool("analyze_certificate", {"target": "example.com"})
        print(len(tools), result)

asyncio.run(main())

Typed async tool functions

The functional API opens a session (a subprocess plus its resilience stack) and passes it to plain async functions:

import asyncio
from dcert_mcp import analyze_certificate, check_expiry, create_session, export_pem

async def main() -> None:
    async with create_session(timeout=60.0) as session:
        cert = await analyze_certificate(target="example.com", session=session)
        expiry = await check_expiry(target="example.com", days=90, session=session)
        pem = await export_pem(target="example.com", output_path="chain.pem", session=session)
        raw = await session.call("tls_connection_info", {"target": "example.com"})

asyncio.run(main())

Leaving out session uses a shared default session that is created on first use and reused afterwards; close_default_session() shuts it down. The shared session belongs to the event loop that created it: a later asyncio.run() gets a fresh one, and calling close_default_session() before a loop ends releases the subprocess cleanly.

Function Description
analyze_certificate() Decode and analyse TLS certificates
check_expiry() Check certificate expiry within N days
check_revocation() Check OCSP revocation status
compare_certificates() Compare certificates between two targets
tls_connection_info() TLS connection details (cipher, protocol, latency)
export_pem() Export the certificate chain as PEM
verify_key_match() Verify a private key matches a certificate
convert_pfx_to_pem() Convert PKCS12/PFX to PEM files
convert_pem_to_pfx() Convert PEM cert and key to PKCS12/PFX
create_keystore() Create a PKCS12 keystore (Java compatible)
create_truststore() Create a PKCS12 truststore from CA certs
tls_scan() Enumerate the TLS versions, cipher suites and groups a server accepts
list_cipher_suites() List TLS cipher suites from the IANA registry (offline)
list_tls_policies() List the catalogued TLS security policies (offline)
show_tls_policy() Show one catalogued TLS security policy (offline)
tls_policy_test() Test an endpoint against a TLS security policy, or identify the closest one
explain_cipher_suite() Explain one cipher suite in any notation (offline)

Every other tool has a wrapper generated from the server's tool schema (dcert_mcp.generated_tools, regenerated by scripts/gen_typed_tools.py and checked in CI): create_csr(), diagnose_endpoint(), explain_edge_term(), validate_certificate(), validate_csr(), vault_issue(), vault_list(), vault_renew(), vault_revoke(), vault_sign(), vault_store(), vault_validate(), verify_key_auto_discover(). Their keyword arguments are the tool's parameters; diagnose_endpoint(timeout_=...) is the tool's connection timeout, because timeout is the per call timeout of every wrapper.

The network facing functions accept the shared mTLS and connection options (client_cert, client_key, pkcs12, cert_password, ca_cert, connect_to, resolve, proxy, noproxy) as keyword arguments.

Error handling

from dcert_mcp import (
    DcertError,            # base class
    DcertTimeoutError,     # the call exceeded its timeout
    DcertConnectionError,  # the subprocess died or the circuit breaker is open
    DcertToolError,        # the tool returned an error result (has .tool)
)

Only transport failures are retried, with exponential backoff and full jitter, and only for tools whose MCP annotations mark them read-only or idempotent (the typed client reads the backend's tool list once per session to decide). TypeError and ValueError surface immediately and cancellation is never swallowed. A call that times out closes the subprocess (it is reconnected on the next call) and counts against the circuit breaker; concurrent calls that fail together share one reconnect and count as one failure.

Command line

dcert-mcp-proxy                                   # stdio (for MCP clients such as Claude Code)
dcert-mcp-proxy --transport http --port 8080      # HTTP; needs DCERT_MCP_AUTH_TOKEN or OIDC
dcert-mcp-proxy --binary /usr/local/bin/dcert-mcp # explicit binary
dcert-mcp-proxy --rate-limit 20 --bulkhead-max 4  # resilience flags
dcert-mcp-proxy --help                            # every option

HTTP and SSE mode refuse to start without authentication: a static bearer token of at least 32 characters (--auth-token or, better, DCERT_MCP_AUTH_TOKEN, compared in constant time) or an OIDC configuration (--oidc-issuer, --oidc-audience, --oidc-jwks-uri or DCERT_MCP_OIDC_ISSUER, DCERT_MCP_OIDC_AUDIENCE, DCERT_MCP_OIDC_JWKS_URL, all https). Requests whose Host header is not a loopback name, the bind address or an entry from --allowed-hosts are answered 400, which defeats DNS rebinding from a browser.

The proxy forwards only the environment variables listed under subprocess.passthrough_env in config.yaml to dcert-mcp: proxy and CA settings, TMPDIR, XDG_CACHE_HOME, DCERT_PROXY, DCERT_NOPROXY, DCERT_KB_FILE, every DCERT_MCP_* policy variable and the VAULT_ADDR, VAULT_TOKEN, VAULT_CACERT, VAULT_CAPATH and VAULT_SKIP_VERIFY settings. An entry ending in * matches by prefix.

dcert-mcp and dcert are the Rust executables themselves, so no Python wrapper sits in front of them.

Configuration

Defaults live in the packaged dcert_mcp/config.yaml: the environment variables forwarded to the subprocess, the server bind address, the transports and exporters offered by dcert-mcp-proxy, and every resilience value. dcert_mcp.load_config() returns them as an immutable record.

Resilience values can be overridden with DCERT_MCP_* environment variables (listed in dcert_mcp/resilience.py) and the dcert-mcp-proxy flags override those. A value outside its valid range (for example DCERT_MCP_CB_THRESHOLD=0 or DCERT_MCP_BULKHEAD_MAX=0) is rejected with a message naming the variable rather than crashing or hanging later. DCERT_MCP_NO_RETRY disables retries in both the proxy and the typed client, and DCERT_MCP_CONNECT_TIMEOUT bounds how long starting the subprocess and reading its tool list may take.

The proxy reads each tool's MCP annotations before applying caching or retries:

  • Caching (--cache) replays results only for tools marked read-only and idempotent. Tools that generate keys, issue or sign certificates, or change Vault state (create_csr, vault_issue, vault_sign, vault_renew, …) are always run.
  • Retry after a transport failure applies only to read-only or idempotent tools. Any other call surfaces the error instead of running a second time, because the first attempt may already have issued a certificate.

Binary discovery

  1. DCERT_MCP_BINARY; a relative path is resolved against the working directory, never through PATH
  2. A compiled dcert-mcp on PATH. The wheel installs it there; pip console script wrappers and the running launcher are skipped.

Nothing is downloaded at run time.

Environment variables forwarded to the binary

The proxy starts dcert-mcp with only these variables from its own environment (subprocess.passthrough_env in config.yaml); everything else is dropped. An entry ending in * forwards every variable with that prefix.

Group Variables
Process HOME, USER, PATH, TMPDIR, XDG_CACHE_HOME
Loader LD_LIBRARY_PATH, DYLD_LIBRARY_PATH, DYLD_FALLBACK_LIBRARY_PATH
Proxy HTTP_PROXY, HTTPS_PROXY, NO_PROXY, ALL_PROXY, http_proxy, https_proxy, no_proxy, all_proxy
Trust SSL_CERT_FILE, SSL_CERT_DIR, REQUESTS_CA_BUNDLE
dcert DCERT_PATH, DCERT_PROXY, DCERT_NOPROXY, DCERT_KB_FILE, DCERT_MCP_*
Vault VAULT_ADDR, VAULT_TOKEN, VAULT_CACERT, VAULT_CAPATH, VAULT_SKIP_VERIFY

DCERT_MCP_* covers the server's own settings, for example DCERT_MCP_TIMEOUT, DCERT_MCP_CONNECTION_TIMEOUT, DCERT_MCP_READ_TIMEOUT and DCERT_MCP_ALLOWED_TARGETS.

Development

The package has no build backend of its own: CI builds the wheels with scripts/build_wheels.py, so pip install -e does not apply. Tests run from the source tree (pythonpath = ["src"] in the pytest configuration) with the extras' requirements installed:

# From the repository root, build the server the integration tests start
cargo build --release --bin dcert-mcp --features mcp
export DCERT_MCP_BINARY="$PWD/target/release/dcert-mcp"

cd python/dcert-mcp
# The requirements of the proxy, otel and dev extras, hash pinned, exactly as CI installs them
pip install --require-hashes -r requirements/dev.txt   # or: uv venv && uv pip install -r requirements/dev.txt

# The same checks as CI
ruff check src tests ../../scripts
ruff format --check src tests ../../scripts
mypy
pytest -q

Without DCERT_MCP_BINARY (or a compiled dcert-mcp on PATH) the integration tests are skipped.

License

MIT

Metadata

Release files for dcert-mcp 4.0.11

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Built distributions (wheels)

Table of built distributions (wheels) for dcert-mcp 4.0.11
File Actions
dcert_mcp-4.0.11-py3-none-win_amd64.whl Python 3 none Windows x86-64 Details for dcert_mcp-4.0.11-py3-none-win_amd64.whl
dcert_mcp-4.0.11-py3-none-manylinux_2_34_x86_64.whl Python 3 none Linux glibc 2.34+ x86-64 Details for dcert_mcp-4.0.11-py3-none-manylinux_2_34_x86_64.whl
dcert_mcp-4.0.11-py3-none-macosx_11_0_arm64.whl Python 3 none macOS 11.0+ ARM64 Details for dcert_mcp-4.0.11-py3-none-macosx_11_0_arm64.whl
dcert_mcp-4.0.11-py3-none-macosx_10_15_x86_64.whl Python 3 none macOS 10.15+ x86-64 Details for dcert_mcp-4.0.11-py3-none-macosx_10_15_x86_64.whl

Total release size: 26.1 MB

Release files / dcert_mcp-4.0.11-py3-none-win_amd64.whl

Download URL dcert_mcp-4.0.11-py3-none-win_amd64.whl
Size 7.4 MB
Tags Python 3 Windows x86-64
SHA-256 checksum
How to use checksums
61c4055faeba605a55dac22eb8ea9977c221b7add2b7976483aa0863397ccc11
BLAKE2b-256 checksum
How to use checksums
17f1af3a94ff74dd2378d01073771bf4b26be5044741ecaaea1921f1fbe50778
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 21, 2026.

Transparency log

Release files / dcert_mcp-4.0.11-py3-none-manylinux_2_34_x86_64.whl

Download URL dcert_mcp-4.0.11-py3-none-manylinux_2_34_x86_64.whl
Size 6.6 MB
Tags Linux glibc 2.34+ x86-64 Python 3
SHA-256 checksum
How to use checksums
6d839db87c05212ab148c9de2702fbf1dca5e62afcede44a1bf21116d70ea4ee
BLAKE2b-256 checksum
How to use checksums
e31af8682b793cc937d73e5c0eb81dd4731849e2f91d0395e92891d5a6ad93c4
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 21, 2026.

Transparency log

Release files / dcert_mcp-4.0.11-py3-none-macosx_11_0_arm64.whl

Download URL dcert_mcp-4.0.11-py3-none-macosx_11_0_arm64.whl
Size 5.9 MB
Tags Python 3 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
a40bd40b5d4a25693508f98ac99f86eb682277f8670050e3d50de79bbeb622f7
BLAKE2b-256 checksum
How to use checksums
73e3a5488f01594e0844344f504607d17c631925d48d52c3acdaa4fd6b062b82
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 21, 2026.

Transparency log

Release files / dcert_mcp-4.0.11-py3-none-macosx_10_15_x86_64.whl

Download URL dcert_mcp-4.0.11-py3-none-macosx_10_15_x86_64.whl
Size 6.2 MB
Tags Python 3 macOS 10.15+ x86-64
SHA-256 checksum
How to use checksums
8e7d31618fff612e2c086f9adc1cbbb52e46da44a0059a41168cda35c7117125
BLAKE2b-256 checksum
How to use checksums
52c0ede3b812ed8487a6f45a3bf43f1766c58a479c2818db875f690f8f0e6e94
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 21, 2026.

Transparency log

Security for dcert-mcp 4.0.11

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 4 files in this release have been uploaded with provenance metadata.

Select from 4 files

Attestation for dcert_mcp-4.0.11-py3-none-win_amd64.whl

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 21, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page