Static worst-case token-budget analysis for LLM-agent workflows + signed budget certificates.
Static worst-case token-budget analysis for LLM-agent workflows. Point it at a Python repo using LangGraph / CrewAI / OpenAI-Agents-SDK and it reports — by pure AST analysis, without running the code — the worst-case budget ceiling of every workflow graph: which units are certifiable / default-dependent / non-certifiable / runaway, and which LLM calls have no token cap. Optionally issues an Ed25519-signed budget certificate logged to a public transparency log. Wraps the hosted costwright API; backed by a Lean 4 cost-soundness theorem.
Use it before deploying an agent workflow to catch missing token caps and
while True:runaway drivers — the budget version of a type check.
| Tool | What it does | Key? |
|---|---|---|
costwright_check(repo_path, policy?) | Static budget analysis of a local repo. Returns pass/fail + counts of certifiable/default-dependent/non-certifiable/runaway units. | yes |
costwright_certify(repo_path, policy?, label?) | Issues a signed, logged budget certificate. Returns cert_id + signature + verify_url. | yes |
costwright_verify(cert_id) | Verify a certificate by id (valid/expired/revoked, signature check). | public |
costwright_pubkey() | Active Ed25519 public keys for offline verification. | public |
{
"mcpServers": {
"costwright": {
"command": "npx",
"args": ["-y", "costwright-mcp"],
"env": { "COSTWRIGHT_API_KEY": "your_rapidapi_key" }
}
}
}
The key is sent as X-RapidAPI-Key (RapidAPI channel) by default; set COSTWRIGHT_DIRECT=1 to send
it as Authorization: Bearer for the direct channel. verify and pubkey work with no key.
check/certify build a .py-only gzip archive of repo_path client-side (excluding venv,
node_modules, tests, etc.) and send it for analysis — your source is uploaded to the hosted API.
See https://eleata.io/privacy/. MIT licensed.
Source-derived launch command. Check the maintainer’s required arguments and credentials before running:
npx -y costwright-mcpMerge this template into ~/Library/Application Support/Claude/claude_desktop_config.json. Keep existing servers. Add any arguments, credentials, and permissions required by the maintainer; this template has not been install-tested.
{
"mcpServers": {
"io-github-hernaninverso-costwright-mcp": {
"command": "npx",
"args": [
"-y",
"costwright-mcp"
]
}
}
}Restart Claude Desktop completely for changes to take effect. Confirm the server appears connected in the client’s tool list, then try a read-only example from its documentation.
Claude Desktop setup referencecostwright — agent budget analysis works with any MCP-compatible client. Copy the config snippet from the Configuration section above and add it to the file shown for your client, then restart the application.
~/Library/Application Support/Claude/claude_desktop_config.jsonRestart Claude Desktop completely for changes to take effect.~/.cursor/mcp.jsonRestart Cursor for changes to take effect..vscode/mcp.jsonReload VS Code window for changes to take effect.~/.codeium/windsurf/mcp_config.jsonRestart Windsurf for changes to take effect..mcp.jsonSave at the project root, then start Claude Code in that project and review the MCP server approval prompt. Keep real credentials out of shared files.