Skip to main content

鵲 - Bridge Claude Desktop to Streamable HTTP MCP servers via mcp-proxy

Project description

cdcasasagi

鵲 - Bridge Claude Desktop to Streamable HTTP MCP servers via mcp-proxy

[!IMPORTANT] As of now, Claude Desktop officially supports custom connectors for remote MCP servers. I recommend considering that option first. While mcp-proxy is a great and easy-to-use tool, the scenarios where it is still necessary are likely limited now that official support exists. Additionally, custom connectors work not only with Claude Desktop but also with the web version of Claude, once configured.

[!NOTE] Windows users: see Windows notes for MSIX-specific config path behavior.

Install

uv tool install cdcasasagi

Usage

Commands at a glance:

  • Mutating (preview by default, --write to apply): add, import, delete
  • Read-only: list, validate-import, doctor
  • Recovery: revert (restores from the .bak left by the last --write), eject (prints managed entries as JSONL and clears them, for re-import after edits)
  • Misc: version

The add, import, and delete commands default to preview mode (no files are modified). Pass --write to apply changes.
For full option details, run cdcasasagi <command> --help.

add

Add a single MCP server entry:

cdcasasagi add https://developers.openai.com/mcp

This shows a unified diff of the proposed change. Pass --write to apply:

cdcasasagi add https://developers.openai.com/mcp --write

A server name is automatically derived from the URL hostname (e.g. developers for the URL above). Use --name to specify a custom name:

cdcasasagi add https://developers.openai.com/mcp --name openai-developer-docs --write

The written entry looks like this:

{
  "mcpServers": {
    "openai-developer-docs": {
      "command": "/Users/you/.local/bin/mcp-proxy",
      "args": [
        "--transport",
        "streamablehttp",
        "https://developers.openai.com/mcp"
      ]
    }
  }
}

import

Add multiple entries at once from a JSONL file:

cdcasasagi import servers.jsonl

Each line is a JSON object with a required url key and optional name / transport keys:

{"url": "https://developers.openai.com/mcp", "name": "openai-developer-docs"}
{"url": "https://example.com/mcp"}

Stdin is also supported — pipe a file, or pass - and paste the JSONL interactively:

cat servers.jsonl | cdcasasagi import -
cdcasasagi import - --write
# Paste JSONL, then press Enter on a blank line to finish
# (Ctrl+D / Ctrl+Z also works)

delete

Remove an entry from the Claude Desktop config by name:

cdcasasagi delete notion

This shows a unified diff of the proposed removal. Pass --write to apply:

cdcasasagi delete notion --write

Names match the left column of cdcasasagi list output. Only entries managed by cdcasasagi (whose command is mcp-proxy) can be deleted; if the named entry was hand-added, delete refuses with an error so you can edit the config manually.

list

Show cdcasasagi-managed MCP servers in the config:

cdcasasagi list

Output is name : url, one entry per line, sorted by name. Only entries whose command is mcp-proxy (or mcp-proxy.exe on Windows) are shown.

validate-import

Validate a JSONL file's schema without importing. This command never writes any files.

cdcasasagi validate-import servers.jsonl

Paste JSONL from stdin instead of preparing a file:

cdcasasagi validate-import -
# Paste JSONL, then press Enter on a blank line to finish
# (Ctrl+D / Ctrl+Z also works)

Once the JSONL validates, feed the same content to import - --write to apply it.

doctor

Check that cdcasasagi can find what it needs to operate:

cdcasasagi doctor

This reports the resolved mcp-proxy binary, the active Claude Desktop config path, and whether the config directory is writable. The command exits with a non-zero status when any check fails, so it can be used in scripts.

On Windows, doctor also surfaces these warnings:

  • Claude Desktop MSIX path — the active config is not under the MSIX virtualized path, but a Claude MSIX install was detected. Claude Desktop on MSIX reads from the virtualized path (%LOCALAPPDATA%\Packages\...\LocalCache\Roaming\Claude\...), so edits to the current path may have no effect. Set CLAUDE_DESKTOP_CONFIG to the candidate path shown in the warning.
  • Orphan APPDATA config — the active config is on the MSIX virtualized path, and a leftover %APPDATA%\Claude\claude_desktop_config.json is also present. Claude Desktop reads only the active file, so any mcpServers entries in the orphan are ignored. Re-add them against the active config, then delete the orphan.
  • Python install path — the Python interpreter that runs cdcasasagi is under %APPDATA%\Roaming, where some Windows security policies block execution. The warning shows a UV_PYTHON_INSTALL_DIR workaround that reinstalls Python under %LOCALAPPDATA%. See #57 for background.

revert

Restore the config from the .bak backup created by the last --write:

cdcasasagi revert

eject

Print all cdcasasagi-managed entries as JSONL on stdout, then remove them from the config. Useful when you want to re-import a slightly modified set of entries: redirect the output to a file, edit it, and re-import.

cdcasasagi eject > backup.jsonl
# edit backup.jsonl
cdcasasagi import backup.jsonl --write

The status summary is printed to stderr so stdout stays a clean JSONL stream that can be redirected or piped. Hand-added entries (whose command is not mcp-proxy) are preserved. Run cdcasasagi revert to undo an eject if needed.

version

cdcasasagi version

Windows notes

cdcasasagi locates Claude Desktop's config in this order:

  1. CLAUDE_DESKTOP_CONFIG environment variable, if set.
  2. The MSIX virtualized path (%LOCALAPPDATA%\Packages\<Claude package>\LocalCache\Roaming\Claude\claude_desktop_config.json), if any MSIX install is detected.
  3. %APPDATA%\Claude\claude_desktop_config.json otherwise.

When multiple MSIX package directories are detected, cdcasasagi first tries to disambiguate by checking which candidates actually contain a claude_desktop_config.json — a common stale-install scenario (several package folders but only one real config) is handled automatically. cdcasasagi only refuses to proceed when more than one candidate file exists and the active one cannot be guessed. In that case, confirm the path via Settings > Developer > Edit Config in Claude Desktop, then set CLAUDE_DESKTOP_CONFIG to that path.

Run cdcasasagi doctor to check the resolved config path. On Windows it also surfaces these situations:

  • The active config is on %APPDATA% while a Claude MSIX install is detected. Claude Desktop on MSIX reads the virtualized path, so %APPDATA% edits may not apply — point CLAUDE_DESKTOP_CONFIG at the MSIX path shown.
  • An orphan %APPDATA%\Claude\claude_desktop_config.json exists alongside the active MSIX config. Claude Desktop ignores the orphan, so any mcpServers entries there are dead. Re-add them against the active config and delete the orphan.
  • The Python interpreter that runs cdcasasagi sits under %APPDATA%\Roaming (uv's default install location). Some Windows security policies block executing binaries from there, which can prevent cdcasasagi and mcp-proxy from running. The warning shows a one-time fix using UV_PYTHON_INSTALL_DIR to install Python under %LOCALAPPDATA% instead. See #57 for the upstream context.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cdcasasagi-0.7.1.tar.gz (15.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

cdcasasagi-0.7.1-py3-none-any.whl (18.8 kB view details)

Uploaded Python 3

File details

Details for the file cdcasasagi-0.7.1.tar.gz.

File metadata

  • Download URL: cdcasasagi-0.7.1.tar.gz
  • Upload date:
  • Size: 15.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.13

File hashes

Hashes for cdcasasagi-0.7.1.tar.gz
Algorithm Hash digest
SHA256 ae112596b3a8dfacf2a3e6d323b00f0b2ea95e420a6f10be699310bb9f6320a9
MD5 c514937990d2280ccf1a8a72b3e3db19
BLAKE2b-256 c051ea4d3db7c87c1cab01f45e586fe1b4dc19565e4969767c04222933bc241e

See more details on using hashes here.

Provenance

The following attestation bundles were made for cdcasasagi-0.7.1.tar.gz:

Publisher: publish.yml on ftnext/cdcasasagi

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file cdcasasagi-0.7.1-py3-none-any.whl.

File metadata

  • Download URL: cdcasasagi-0.7.1-py3-none-any.whl
  • Upload date:
  • Size: 18.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.13

File hashes

Hashes for cdcasasagi-0.7.1-py3-none-any.whl
Algorithm Hash digest
SHA256 883969cae1adcfeb0608909c9f56a708d809a91a18bfe352239811fad5bd4ef5
MD5 b644f3681ba2f9fec0d7f000214d10f2
BLAKE2b-256 b09e8d03deb8146470345e37dd7632eec45b4b1bfb6b27939fe2d96bc2d23cff

See more details on using hashes here.

Provenance

The following attestation bundles were made for cdcasasagi-0.7.1-py3-none-any.whl:

Publisher: publish.yml on ftnext/cdcasasagi

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page