Gib deinem KI-Agenten das Eine, das er nicht selbst tun kann: Code tatsächlich gegen Tests auszuführen und zu beweisen, dass er bestanden hat.
Soma ist ein Code-Dienst mit Ausführungsverifizierung. Dieser MCP-Server stellt zwei Werkzeuge bereit:
soma_verify_code — führe Kandidatencode gegen Tests in einer isolierten Sandbox aus; erhalte ein PASS/FAIL-Urteil plus ein signiertes, offline prüfbares Zertifikat (Ed25519). Nutze es, um unabhängig zu bestätigen, dass der Code funktioniert, bevor du ihm vertraust.
soma_generate_verified_code — bitte Soma, Code für eine Aufgabe zu schreiben; wenn die Aufgabe verifizierbar ist, wurde der zurückgegebene Code bereits gegen abgeleitete Tests ausgeführt und mit einem Zertifikat versehen.
15+ Sprachen werden für die Verifizierung unterstützt (Python, JavaScript/TypeScript, Go, C/C++, Java, Rust, Ruby, PHP, Bash und weitere).
Erfordert Node.js 18+. Läuft über stdio.
Füge es zu deiner MCP-Client-Konfiguration hinzu (Claude Desktop, Cursor usw.):
{
"mcpServers": {
"soma": {
"command": "npx",
"args": ["-y", "soma-verify-mcp"],
"env": {
"SOMA_API_KEY": "YOUR_SOMA_KEY"
}
}
}
}Claude Desktop: Settings → Developer → Edit Config, füge den obigen Block hinzu und starte neu.
Cursor: Settings → MCP → Add, oder füge denselben Block in ~/.cursor/mcp.json ein.
Related MCP server: OmniBridge MCP Server
Umgebungsvariable | Erforderlich | Standard | Zweck |
| ja | — | Dein Soma-API-Schlüssel. |
| nein |
| Basis-URL der Soma-API. |
| nein |
| Timeout pro Anfrage. |
Erhalte einen kostenlosen Vorschau-Schlüssel: Kontaktiere centrum.arvind@gmail.com (kostenloses Kontingent während der Vorschau).
soma_verify_codeFühre Code gegen Tests aus und gib ein signiertes Urteil zurück.
language (string) — z. B. python, javascript, go, rust.
code (string) — der vollständige Quellcode, der verifiziert werden soll.
tests (array) — eines von:
Funktionsmodus (Standard): [{ "input": [arg1, arg2], "expected": value }] plus entrypoint (der Funktionsname).
Stdio-Modus: setze mode: "stdio" und [{ "stdin": "...", "expected_stdout": "..." }]; kein entrypoint.
entrypoint (string, optional) — Funktionsname für den Funktionsmodus.
mode ("function" | "stdio", optional).
Gibt zurück: verdict, tests_passed, tests_total sowie eine signature / public_key / sig_alg, die du offline prüfen kannst.
soma_generate_verified_codeErhalte Code für eine Aufgabe, der vor der Rückgabe gegen abgeleitete Tests ausgeführt wurde.
prompt (string) — die Programmieraufgabe. Füge konkrete Eingabe-/Ausgabebeispiele hinzu (z. B. >>> f(2) == 4), damit das Ergebnis verifizierbar ist und nicht nur nach bestem Bemühen.
max_tokens (int, optional, Standard 1500).
Gibt zurück: den Code, certified (bool) und ein certificate (verdict, tests_passed, tests_total), wenn die Verifizierung bestanden wurde. Wenn eine Aufgabe nicht verifizierbar ist, wird die Ausgabe ohne Zertifizierung und klar gekennzeichnet zurückgegeben — niemals ein falsches „verified“.
Ein Zertifikat bestätigt, dass die aufgeführten Tests zum Zeitpunkt der Erstellung in einer isolierten Sandbox bestanden wurden. Es ist signiert (Ed25519) und offline anhand des zurückgegebenen öffentlichen Schlüssels prüfbar. Es ist keine Garantie für die Eignung für irgendeinen Zweck — überprüfe die Ausgabe vor dem Produktionseinsatz.
Kein Training mit deinen Prompts. Siehe die Soma-Datenschutz- und Datenrichtlinie unter ${SOMA_BASE_URL}/privacy.
MIT.
Ask Soma to write code for a task. When the task is verifiable, the returned code has been executed against derived tests in an isolated sandbox before it is returned, and a certificate (verdict + tests passed) is attached. Include concrete examples in the prompt (e.g. doctest-style '>>> f(2) == 4') to make the result verifiable rather than best-effort.
| Name | Required | Description | Default |
|---|---|---|---|
| prompt | Yes | The coding task. Include example input/output pairs to enable verification. | |
| max_tokens | No | Maximum output tokens (default 1500). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and destructiveHint=false, indicating a safe, non-destructive operation. The description adds valuable behavioral context: the code is executed in an isolated sandbox, a certificate with verdict and tests passed is attached, and the verification is conditional on task verifiability. This goes beyond the annotations and provides important expectations about the output and process.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise and well-structured. It front-loads the core purpose, then explains the verification process and provides actionable advice on prompt construction. Every sentence adds value, and there is no fluff or repetition. The structure is logical: what it does, how it works, and how to use it effectively.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's complexity (verification, sandbox, certificate) and the absence of an output schema, the description covers the essential aspects: what the tool does, how verification works, and how to improve results. It does not detail the certificate format or the exact conditions for verifiability, but these are not critical for an agent to invoke the tool correctly. The description is complete enough for effective use.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so both parameters (prompt and max_tokens) are already documented in the schema. The description adds guidance on how to use the prompt parameter (include concrete examples) and mentions the default for max_tokens, but this is largely redundant with the schema. The description does not add significant new meaning beyond the schema, so a baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose: 'Ask Soma to write code for a task.' It specifies the key differentiator—execution-verified code with a certificate—and distinguishes it from a best-effort generation. The verb 'write code' and resource 'task' are specific, and the mention of verification sets it apart from the sibling tool.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear guidance on when to use this tool: when the task is verifiable, and it advises including concrete examples in the prompt to enable verification. It implies that for non-verifiable tasks, this tool may not be ideal, but it does not explicitly name the sibling tool or state when to use soma_verify_code instead. The guidance is strong but lacks an explicit exclusion or alternative mention.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Run candidate code against tests inside an isolated sandbox and return a PASS/FAIL verdict with a signed, offline-checkable certificate (Ed25519). Use this to independently confirm that code actually works before trusting it.
Two test shapes:
function mode (default): tests = [{"input": [arg1, arg2], "expected": value}] and provide 'entrypoint' (the function name).
stdio mode: set mode='stdio' and tests = [{"stdin": "...", "expected_stdout": "..."}]; no entrypoint needed.
Supported languages include python, javascript, typescript, go, c, cpp, java, rust, ruby, php, bash and more.
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes | The complete source code to verify. | |
| mode | No | 'function' (default) or 'stdio'. | |
| tests | Yes | function mode: [{input:[args], expected: value}]. stdio mode: [{stdin:'...', expected_stdout:'...'}]. | |
| language | Yes | Programming language, e.g. 'python', 'javascript', 'go', 'rust'. | |
| entrypoint | No | Function name to call (required for function mode; omit for stdio mode). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description reveals the isolated sandbox behavior and the signed certificate output, adding value beyond the annotations. It does not specify sandbox limits, timeouts, or what happens on ambiguous test results, but the key isolation and non-destructive nature are explicit.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Well-organized with clear separation of modes, languages, and purpose. Slightly verbose but every line adds value. Could be tightened, but readability is strong.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Covers when to use (independent confirmation), the two modes, supported languages, and the output certificate nature. Missing: no example usage, no details on error handling or timeouts, and output format isn't specified, but the description is otherwise informative enough for an agent to call it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema documents all five parameters with types and required status, and the description adds concrete examples for tests structures in both modes. Entry point requirement is clarified. Coverage is 100% and the semantics are unambiguous.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool verifies code against tests in a sandbox and returns a signed certificateat. Uses specific verbs (verify, return PASS/FAIL) and names two precise modes. Though siblings weren't supplied for comparison, it self-defines scope well.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides clear usage context: 'Use this to independently confirm that code actually works before trusting it.' Describes two modes (function and stdio) with examples. Does not explicitly list when-not-to-use or alternatives, but the context is enough for most agents.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0soma_generate_verified_codesoma_verify_codeScored across 2 tools
The two tools have clearly distinct purposes: one verifies provided code, the other generates code with verification. There is no overlap or confusion between them.
Both tools follow the identical pattern of 'soma_' prefix plus a verb_noun structure: verify_code and generate_verified_code. Naming is uniform and predictable.
With only 2 tools, the server is minimal but focused on a narrow purpose. While the count feels thin for a general toolkit, it is reasonable for a specialized verification service. It sits at the borderline.
The server covers the core workflows of code verification and generation with verification. Minor gaps exist, such as no ability to fetch or check certificate details after generation, but the primary lifecycle is covered.
Hosted Python sandbox for package runtime verification: pip install and usable-import evidence.
Execute code in 8 languages (Python, JS, TS, Go, Java, C++, C, Bash) in gVisor sandboxes.
Signed sandbox verdicts on any artifact, plus an agent marketplace. No key, no signup, no payment.
Attestation infrastructure for the agentic economy: signed, independently verifiable verdicts.
