Missions offener MCP-Interceptionsadapter für exakte Aktionsautorität.
Es hat eine einzige, enge Aufgabe: Vor einem gewrappten MCP-Anbieter zu sitzen, einen vorgeschlagenen Tool-Aufruf zu klassifizieren und folgenreiche Aufrufe mit einer unveränderlichen Trust Graduation-Aktionsbindung zurückzuhalten. Risikoarme Aufrufe können durchgelassen werden. Eine Chat-Nachricht, die „approve“ sagt, wird niemals als Autorität behandelt.
Status: experimentelle Beta. Apache-2.0. Keine Laufzeitabhängigkeiten.
Anforderungen: Node.js 20 oder neuer.
npx -y @gomission/mcp@beta demoDer Befehl verwendet einen gefälschten E-Mail-Anbieter und gibt ein maschinenlesbares DEMO_RESULT aus. Er beweist, dass:
die folgenreiche Provider-Funktion nullmal aufgerufen wurde;
der Hold eine exakte Aktion und einen Input-Hash enthält;
das Ändern des überprüften Inputs die Bindung (Commitment) ändert;
Chat-Text keine Ausführungsautorität gewährt.
Dies beweist die Interception, nicht die abgeschlossene Autorisierung oder Produktionsübernahme. Der portable Grant/Replay-Beweis befindet sich in @trust-graduation/core:
npx -y @trust-graduation/core@beta demoRelated MCP server: agentpassport
Primitive | Verantwortung | Tragbares Objekt |
Mission Gate | Entscheide vor einem Provider-Effekt | Aktionsentscheidung |
Trust Profile | Verfolge erworbene Autorität pro Principal und Aktionsklasse | Nachweisprofil |
Mission Key | Autorisiere eine exakte Aktion bis Ablauf oder Widerruf | Einmaliger Grant |
Der MCP-Adapter implementiert den Pre-Provider-Hold. Er prägt keinen vertrauenswürdigen Mission Key und kann einen zurückgehaltenen Aufruf nicht wieder aufnehmen. Ein vertrauenswürdiger Approval-Host und Executor müssen den passenden Schlüssel validieren und atomar verbrauchen. Die experimentelle A2A-Fortsetzung ist veröffentlicht unter:
https://trustgraduation.org/extensions/a2a/action-authorization/v1
Das Paket-Root stellt die stabile, abhängigkeitsfreie Binding-Brücke bereit, die von einem externen Approval-Host und Executor verwendet wird:
import { providerActionFromMcpBinding } from "@gomission/mcp";
import { createProviderGate } from "@trust-graduation/core";
const gate = createProviderGate({
store: sharedAtomicGrantStore,
authenticateGrant: verifyApprovalIssuer,
provider: existingProviderFunction,
writeReceipt: durableReceiptSink
});
// Re-read the actual provider input at the final seam; never trust a preview.
const action = providerActionFromMcpBinding(
heldReceipt.action_binding,
actualProviderInput
);
const execution = await gate.execute({
binding: heldReceipt.action_binding,
approval: authenticatedMissionKey,
action
});Die Brücke überprüft die Integrität der Bindung und bildet die abgefangenen Identitäten, das Ziel, die Einschränkungen, den Ablauf und die Nonce auf die Form des Core-Executors ab. Der Core hasht dann den tatsächlichen Provider-Input erneut, authentifiziert und verbraucht den Key atomar, ruft den Provider auf und schreibt ergebnisverknüpfte Nachweise. Mutation oder Wiedergabe erreichen den Provider nie.
Für einen generierten Adapter und objektive Provider-Call-Zähler:
npm install @trust-graduation/core@beta
npx trust-graduation init-adapter
npx trust-graduation conformance ./mission-gate-adapter.mjs --jsonMit beiden installierten Paketen lautet der enthaltene Kompatibilitätsbeweis:
node node_modules/@trust-graduation/core/examples/mcp-provider-roundtrip.mjsDer MCP-Proxy nimmt einen zurückgehaltenen Aufruf immer noch nicht wieder auf, nur weil der Chat „approve“ sagt. Diese Brücke ist für den separat authentifizierten Approval-Host und den providergebundenen Executor.
npx -y @gomission/mcp@beta install-claudeDer Installer überprüft die vorhandene Claude Desktop MCP-Konfiguration:
wenn es folgenreiche MCP-Server findet, wählt es --wrap;
andernfalls wählt es --local, eine beratende exakte Bindungsdemonstration;
es wählt niemals automatisch den gehosteten Nur-Lese-Modus.
Starten Sie Claude Desktop nach der Installation neu und überprüfen Sie dann:
npx -y @gomission/mcp@beta verifyverify testet modernes MCP mit server/discover und tools/list und fällt auf das Initialize-Protokoll für ältere Endpunkte zurück. Fügen Sie --json für einen maschinenlesbaren Bericht oder --no-probe hinzu, um nur die Konfiguration zu überprüfen.
Modus | Was es erzwingt | Was es nicht tut |
| Fängt ausgewählte Child-MCP-Server ab; hält hohe/kritische oder niedrig vertrauenswürdige Aufrufe vor dem Child zurück; schließt fehl, wenn ein Child nicht verfügbar ist | Nimmt keinen zurückgehaltenen Aufruf wieder auf oder vertraut Chat-Genehmigung |
| Zeichnet einen beratenden exakten Aktions-Hold und eine lokale Überprüfungsquittung auf | Ist nicht zwischen einem anderen Tool und seinem Provider |
| Stellt gehosteten schreibgeschützten Mission-Kontext bereit | Fängt keine anderen MCP-Server ab |
Wählen Sie explizit, wenn nötig:
npx -y @gomission/mcp@beta install-claude --wrap
npx -y @gomission/mcp@beta install-claude --local
npx -y @gomission/mcp@beta install-claude --remoteNützliche Flags:
--workspace <path> — speichert lokale Quittungen in einem vorhandenen Workspace.
--dry-run — gibt die Konfigurationsänderung aus, ohne sie zu schreiben.
--force — erstellt Konfiguration, auch wenn Claude Desktop nicht erkannt wird.
--remote-url <url> — überschreibt den gehosteten Endpunkt.
MISSION_DONT_WRAP="name1,name2" — schließt ausgewählte MCP-Childs aus.
Für einen folgenreichen gewrappten Aufruf schreibt der Adapter eine lokale Quittung mit folgenden Inhalten:
Aktionsklasse;
datenschutzschonende lokale Workspace-Kennung;
anforderndes MCP-Child;
Ziel, wenn eines abgeleitet werden kann;
SHA-256 Input-Commitment;
Einmalausführungsbeschränkungen;
Ablauf und Nonce;
SHA-256-Commitment über die gesamte Bindung.
Quittungen werden atomar mit dateibesitzerbezogenen Berechtigungen geschrieben. Der Adapter speichert niemals den rohen Workspace-Pfad innerhalb der Bindung. Eine lokale Argumentzusammenfassung verbleibt in der Quittung zur menschlichen Überprüfung, behandeln Sie daher das Quittungsverzeichnis als sensible Workspace-Daten.
Bevorzugtes Protokoll: 2026-07-28.
zustandsloses _meta pro Anfrage mit Client-Fähigkeiten;
obligatorisches server/discover;
eine JSON-RPC-Nachricht pro HTTP-POST; moderne Batches und Client-Benachrichtigungen schlagen fehl;
HTTP-Bindung für MCP-Protocol-Version, Mcp-Method und Mcp-Name;
protokolldefinierte HeaderMismatch- und nicht unterstützte Versionsfehler;
resultType: "complete" und Cache-Metadaten;
Initialize-Ära-Kompatibilität für 2025-11-25 und 2024-11-05.
Das Autoritätsmanifest wird über MCP-Discovery unter der experimentellen mission-authority/v1-Fähigkeit beworben.
Der Adapter beansprucht nicht:
dass ein Modell oder eine Chat-Benutzeroberfläche den Principal authentifiziert hat;
dass eine Überprüfungsquittung ein Genehmigungsgrant ist;
Exactly-Once-Verhalten bei einem externen Provider;
unabhängige Konformitäts- oder Produktionsvalidierung;
globales Vertrauen in einen Agenten.
Verwenden Sie @trust-graduation/core, um exakte Grants zu erstellen und zu validieren. Der Executor muss den Grant-Aussteller authentifizieren, den tatsächlichen Provider-Input neu binden, den Schlüssel atomar verbrauchen, den Provider höchstens einmal aufrufen und unbekannte Provider-Ergebnisse abgleichen.
Frei und offen:
dieser MCP-Adapter;
@trust-graduation/core und seine Schemata;
@gomission/mission-schemas Konformitätsvokabular;
die A2A-Erweiterung für exakte Aktionsautorisierung;
Mission Lites lokale Fokus-App.
Kommerzielles Mission kann verwaltete Richtlinien, vertrauenswürdige Genehmigungsoberflächen, gehostete Audit-/Quittungsvorgänge, Organisationskontrollen, Support und Provider-Integrationen bereitstellen. Produktberechtigungen gewähren niemals Aktionsautorität.
Protokoll: https://trustgraduation.org/
A2A-Erweiterung: https://trustgraduation.org/extensions/a2a/action-authorization/v1
Mission: https://gomission.io/
Quelle: https://github.com/gomission/mcp
Apache-2.0. Missionsnamen und -logos sind Marken; die Codelizenz gewährt keine Erlaubnis, eine Billigung zu implizieren.
Fetch a Mission receipt by id.
| Name | Required | Description | Default |
|---|---|---|---|
| receipt_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, and the description only states a basic fetch operation without disclosing any behavioral traits such as read-only nature, error handling, or performance implications.
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 a single concise sentence with no waste, but it could benefit from a bit more detail without losing brevity.
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 simplicity of the tool (one parameter, no output schema, no annotations), the description is minimally adequate but lacks information on return format, error conditions, or how receipt_id is obtained.
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 input schema has one parameter (receipt_id) with no description, and the tool description only mentions 'by id' without explaining format, examples, or how to obtain the id.
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 'Fetch a Mission receipt by id' uses a specific verb and resource, clearly distinguishing it from sibling tools like log_action or mission_status.
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 no guidance on when to use this tool versus alternatives, and no exclusions or context are given.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Log a safe internal action with Mission. Returns a receipt id. Use for non-external actions that still benefit from a permanent receipt.
| Name | Required | Description | Default |
|---|---|---|---|
| summary | Yes | ||
| evidence | No | ||
| action_class | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description must fully disclose behavior. It only states 'safe internal action' and output, but omits details about authentication, side effects, or error conditions. The term 'safe' is vague and insufficient.
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?
Very concise: two sentences, no filler. Key information is front-loaded. Every sentence adds value.
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 has 3 parameters, no output schema, and no annotations, the description is too sparse. It provides no parameter guidance, no return value details beyond receipt id, and no error handling. The complexity is not matched by the description.
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 coverage is 0% (no parameter descriptions in schema). The description does not mention any parameters, their purpose, or constraints. It fails to compensate for the lack of schema documentation.
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?
Description clearly states action 'log a safe internal action' and output 'returns a receipt id'. It gives a specific verb and resource, but does not explicitly differentiate from siblings like 'get_receipt' or 'request_approval', though the internal purpose is implied.
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 guidance on when to use: 'Use for non-external actions that still benefit from a permanent receipt.' This gives a clear context, but lacks explicit exclusions or alternatives among siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Check the explicitly selected local Mission workspace for chat availability. Requires MISSION_WORKSPACE and an explicit loopback MISSION_LOCAL_URL or MISSION_LOCAL_PORT. Chat is currently unavailable until a workspace-scoped authentication handoff is supported; this tool does not forward queries or prepare actions.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | A natural-language question or instruction for Mission. Examples: 'what should I approve today?', 'who am I dropping?', 'prepare a reply to Sandro', 'show me my weekly proof', 'what did I just learn?' |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and delivers unusually candid behavior: it discloses required environment variables and, critically, that chat is currently unavailable pending a workspace-scoped authentication handoff. It also states the tool has no side effects (no query forwarding, no action preparation). It falls short of a 5 only because it does not describe what the call actually returns in its unavailable state.
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?
Three sentences with the core purpose front-loaded and no obvious padding. It is dense but efficient; minor jargon ('loopback', 'workspace-scoped authentication handoff') slightly slows comprehension.
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 is a non-functional availability probe with no output schema, the description is largely complete: it covers required env configuration, the blocked state, and the absence of side effects. It would be fully complete if it pointed the agent at mission_status or otherwise told it what to do instead.
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% and the single required 'query' parameter is fully documented with examples in the schema, so the baseline is 3. The line 'does not forward queries' does tell the agent the parameter is effectively inert, which is useful, but otherwise the description adds no syntax or format detail beyond the schema.
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 states a specific action and scope: checking the explicitly selected local Mission workspace for chat availability, and explicitly narrows it by saying it 'does not forward queries or prepare actions.' This distinguishes it from request_approval and log_action. It loses a point because the name 'mission_ask' implies a working Q&A tool, creating tension the description only partially resolves, and it never contrasts itself with mission_status.
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?
Preconditions are stated clearly (MISSION_WORKSPACE plus an explicit loopback MISSION_LOCAL_URL or MISSION_LOCAL_PORT), which is real context. However, there is no explicit when-to-use vs when-not-to-use guidance relative to siblings like mission_status, and the reader must infer that this tool is effectively a no-op probe.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Return Mission gate status, version, and the current Trust Graduation action classes. Call this first in any session to introduce the permission layer.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It describes the return values but does not explicitly state that it has no side effects or that it is read-only. However, the nature of returning status information implies safety, so it is adequate but not exceptional.
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?
Two concise sentences: the first states what it returns, the second gives usage guidance. No unnecessary words, highly efficient.
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 no output schema, the description adequately lists the three return categories (gate status, version, action classes). It could provide more detail on the format, but it is sufficient for a simple tool with no parameters.
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 tool has zero parameters, so baseline is 4. The description does not need to add parameter meaning; it correctly omits parameter information.
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 returns 'Mission gate status, version, and the current Trust Graduation action classes'. It distinguishes itself from siblings by instructing to call it first in any session to introduce the permission layer, making its unique role obvious.
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?
Explicitly tells when to use: 'Call this first in any session'. This implies it is a prerequisite for other tools, though it does not explicitly state when not to use it or provide alternatives.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Legacy-named advisory hold. Call before a consequential action to produce an exact action binding and local review receipt. This tool does not grant authority or resume execution; the executor must validate a matching single-use Mission Key.
| Name | Required | Description | Default |
|---|---|---|---|
| summary | Yes | One-line plain-English summary of the action. | |
| evidence | No | Optional evidence supporting the action. | |
| action_class | Yes | One of: send_email, post_public, send_dm, schedule_meeting, spend_money, publish_artifact, modify_external_record, change_trust_policy |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and discharges much of it: it discloses that the call is advisory-only, non-authoritative, does not resume execution, and that the executor must validate a matching single-use Mission Key. It omits auth requirements and what the receipt actually contains, which are notable gaps for a zero-annotation tool.
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?
Three sentences with no filler, and the most consequential caveat (legacy name, advisory-only) is front-loaded before the invocation condition. Every sentence changes how an agent would behave.
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?
For a 3-parameter, no-annotation, no-output-schema tool, the description covers the critical semantics: advisory scope, non-authority, and the Mission Key handoff. It does not describe the receipt's contents or the full Mission Key flow, which leaves a small but real gap.
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 action_class, summary, and evidence are already documented with their allowed values. The description adds no syntax or format guidance beyond the schema, so the 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 names a concrete deliverable (an exact action binding and a local review receipt) for a consequential-action workflow, and flags that the tool name 'request_approval' is legacy and misleading. It does not, however, position itself against siblings like mission_ask or get_receipt, so the agent gets the function but not the sibling routing.
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?
It gives clear positive timing ('Call before a consequential action') and an explicit negative boundary ('does not grant authority or resume execution'). It stops short of naming which sibling to use instead when the agent actually needs authority or status, so the alternative-routing half of the guidance is missing.
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.
v0.1.0get_receiptlog_actionmission_askmission_statusrequest_approvalScored across 5 tools
Each tool targets a distinct action: request_approval for consequential holds, log_action for internal receipts, get_receipt for retrieval, mission_status for gate status, and mission_ask for chat availability. The overlap between request_approval and log_action (both producing receipts) is clarified by descriptions, but the legacy advisory nature of request_approval could still cause confusion.
Most tools use a verb_noun pattern (request_approval, get_receipt, log_action), but mission_ask and mission_status use a noun-prefix convention, and request_approval is explicitly labeled legacy-named. The mix is readable but not fully consistent.
Five tools is a reasonable size for a focused permission-layer server. However, mission_ask is currently non-functional and request_approval is legacy, so one tool does not fully earn its place, making the count slightly heavy rather than ideal.
The core lifecycle of status, approval request, action logging, and receipt retrieval is covered, but there are notable gaps: no way to list or search receipts, no tool to validate the single-use Mission Key that request_approval requires the executor to check, and mission_ask only checks unavailable chat functionality.
Decision-assurance for AI agents: an auditable action boundary + receipt before it acts.
Trust gate for AI agents: multi-model adversarial consensus, signed and verifiable verdicts.
Independent effect verification and signed receipts for consequential AI agent actions.
Pre-execution policy gate for consequential agent actions with durable trust receipts.
