MCP server

Bytecode recovery, available to your agent.

Connect a compatible AI client to one authenticated Streamable HTTP endpoint. PyDecode handles credits, upload validation, isolated decompilation, job polling, and bounded source delivery.

Create a key Download config

Remote

No local server to maintain

Private

Bearer-scoped jobs and results

Compatible

2026 protocol + 2025 fallback

Connect

One endpoint, one revocable key.

Use a dedicated PyDecode API key as a bearer token. Client configuration formats vary, but every remote connection needs the URL and Authorization header shown below.

{
  "mcpServers": {
    "pydecode": {
      "type": "http",
      "url": "https://pydecode.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_PYDECODE_API_KEY>"
      }
    }
  }
}

Tool reference

A complete recovery workflow.

Tools are intentionally narrow so clients can reason about credit use, retries, polling, and untrusted recovered source.

Read only

check_credits

Returns the balance and recent immutable ledger entries.

Uses credits

submit_decompilation

Accepts one authorized .pyc or ZIP as standard base64 and queues static recovery.

Read only

get_job

Returns state, credit reconciliation, output counts, timestamps, and events.

Read only

list_jobs

Lists private jobs newest first with cursor pagination.

Read only

read_result

Returns bounded recovered .py source from completed jobs for agent review.

How an agent uses it

Submission is asynchronous so long-running recovery never holds an MCP request open.

1Check the balance before a large archive.
2Encode the authorized .pyc or ZIP and submit it.
3Poll the returned private job ID until completion.
4Read bounded recovered source or download the full ZIP over REST.

Ready to connect?

Create a dedicated MCP key.

Keys are independently revocable. Uploaded bytecode is never executed, and returned source must always be treated as untrusted code.

Open MCP access