AssistantHub MCP API Reference

AssistantHub.McpServer is a standalone Voltaic-based MCP server that exposes the AssistantHub management and configuration surface as MCP tools. It connects to an upstream AssistantHub REST server with the C# SDK and uses a small REST proxy for exact REST pass-through cases such as configuration replacement, binary download wrappers, and HEAD-style existence checks.

This document is the source of truth for the MCP surface. For the underlying HTTP API, see REST_API.md. For Claude/Cursor setup guidance, see docs/CLAUDE_MCP.md.

Transports

Default transports:

Transport	Default endpoint	Notes
HTTP JSON-RPC	`http://127.0.0.1:8820/rpc`	Voltaic HTTP transport
HTTP events	`http://127.0.0.1:8820/events`	Used by the HTTP MCP transport
TCP	`tcp://127.0.0.1:8821`	Voltaic TCP transport
WebSocket	`ws://127.0.0.1:8822/mcp`	Voltaic WebSocket transport

Container defaults use 0.0.0.0 for the bind host so the ports are reachable outside the container.

Quick Start

Build the solution:

dotnet build src/AssistantHub.sln

Run AssistantHub Server:

dotnet run --project src/AssistantHub.Server/AssistantHub.Server.csproj

Run the MCP server:

dotnet run --project src/AssistantHub.McpServer/AssistantHub.McpServer.csproj

Preview the generated MCP configuration:

dotnet run --project src/AssistantHub.McpServer/AssistantHub.McpServer.csproj -- --showconfig

Install Claude/Cursor snippets from the built MCP server:

cd src/AssistantHub.McpServer/bin/Debug/net10.0
./AssistantHub.McpServer install --dry-run
./AssistantHub.McpServer install

Configuration

Default config file: assistanthub-mcp.json

Example:

{
  "AssistantHub": {
    "Endpoint": "http://localhost:8800",
    "ApiKey": "default"
  },
  "Http": {
    "Hostname": "127.0.0.1",
    "Port": 8820
  },
  "Tcp": {
    "Address": "127.0.0.1",
    "Port": 8821
  },
  "WebSocket": {
    "Hostname": "127.0.0.1",
    "Port": 8822
  },
  "Storage": {
    "BackupsDirectory": "./backups/",
    "TempDirectory": "./temp/",
    "MaxInlineBinaryBytes": 5242880
  }
}

Supported environment overrides:

Variable	Purpose
`ASSISTANTHUB_ENDPOINT`	Upstream AssistantHub server URL
`ASSISTANTHUB_API_KEY`	Upstream bearer token or admin API key
`MCP_HTTP_HOSTNAME`	HTTP bind host
`MCP_HTTP_PORT`	HTTP bind port
`MCP_TCP_ADDRESS`	TCP bind address
`MCP_TCP_PORT`	TCP bind port
`MCP_WS_HOSTNAME`	WebSocket bind host
`MCP_WS_PORT`	WebSocket bind port
`MCP_CONSOLE_LOGGING`	Console logging toggle

Tool Naming Rules

Tools use lowercase slash-delimited names.
CRUD-style tools follow domain/action.
Sub-resource tools follow domain/subdomain/action.
HTTP HEAD existence checks are normalized as */exists.
Public assistant metadata helpers remain under the assistant/* namespace.

Examples:

tenant/create
assistant/settings/get
assistantanalytics/overview
bucket/object/upload
requesthistory/summary
eval/judge-prompt/default

Secret Handling

Redaction is on by default for responses serialized through the MCP helper layer. The redactor masks these property names case-insensitively:

BearerToken
Password
AdminPassword
DefaultAdminPassword
AdminApiKeys
ApiKey
ApiKeyValue
AccessKey
SecretKey
SlackAppToken
SlackBotToken

Important behaviors:

configuration/get returns a redacted configuration by default.
assistant/settings/get and assistant/settings/update return redacted secret-bearing fields by default.
credential/list, credential/get, credential/create, and credential/update redact bearer tokens by default.
Set includeSecrets=true only when the caller explicitly needs the raw secret values.

Binary And Streaming Rules

Binary wrappers:

document/upload
document/download
bucket/object/upload
bucket/object/download

Binary contract:

uploads use contentBase64
downloads return FileName, ContentType, Size, ContentBase64, and Source
default inline size limit is 5242880 bytes (Storage.MaxInlineBinaryBytes)
oversized binary payloads return an error instead of an external URL

Streaming status:

Eval SSE is not exposed through MCP in this release.
Public assistant chat/generate/compact/feedback/download flows are not exposed through MCP in this release.
Assistant public document listing is exposed as assistant/documents/list and mirrors GET /v1.0/assistants/{assistantId}/documents. Sending attached-document chat requests remains REST-only in this release; send attached_document_ids to POST /v1.0/assistants/{assistantId}/chat.
Model-directed runtime tools such as collection search, Verbex search, S3 object reads, and Tavily web search are not exposed as MCP tools for public chat users. Configure assistant tool policy, optional ToolRoutingInferenceEndpointId, and optional default-off ExposeThinking through the REST API, SDKs, or dashboard. Admins can inspect redacted global Tavily readiness through GET /v1.0/configuration/external-search/status or the SDK status helpers. When enabled, REST assistant chat executes those tools server-side against explicit OpenAI-compatible or Ollama tool-capable completion endpoints; a dedicated tool-routing endpoint may decide tool calls while InferenceEndpointId still writes final answers. Streaming REST chat can emit safe tool-progress SSE events, including heartbeat events for long-running tool calls, and can emit provider thinking deltas only when the assistant permits it, while MCP remains management-only for this release.
Use the REST API directly for these streaming or interaction-heavy routes.

Tool Families

Family	Representative tools
System	`system/health`, `system/whoami`, `system/openapi`
Authentication	`auth/authenticate`
Tenants / Users / Credentials	`tenant/`, `user/`, `credential/*`
Storage	`bucket/`, `bucket/object/`
Collections	`collection/`, `collection/record/`, `collection/search`
Indices	`index/`, `index/record/`, `index/search`
Assistants	`assistant/`, `assistant/settings/`, `assistant/tool-calls/*`
Documents / Ingestion	`document/`, `ingestionrule/`
Monitoring	`history/`, `thread/`, `requesthistory/`, `assistantanalytics/`
Endpoint management	`embeddingendpoint/`, `completionendpoint/`, `model/*`
Crawl	`crawlplan/`, `crawloperation/`
Evaluation	`eval/fact/`, `eval/run/`, `eval/result/get`, `eval/judge-prompt/default`
Runtime configuration	`configuration/get`, `configuration/update`

Crawl Plan Repository Types

The crawlplan/create and crawlplan/update tools accept a planJson string containing the same CrawlPlan JSON contract used by REST and the SDKs. Supported RepositoryType values are:

Value	Settings type	Notes
`Web`	`WebCrawlRepositorySettings`	Uses `StartUrl`, web authentication fields, link-following settings, sitemap/robots settings, depth, parallelism, and crawl delay.
`CIFS`	`CifsCrawlRepositorySettings`	Uses `CifsHostname`, `CifsUsername`, `CifsPassword`, `CifsShareName`, and `IncludeSubdirectories`.
`NFS`	`NfsCrawlRepositorySettings`	Uses `NfsHostname`, `NfsUserId`, `NfsGroupId`, `NfsShareName`, `NfsVersion`, and `IncludeSubdirectories`.

CIFS passwords, web passwords, bearer tokens, and API keys are secret-bearing repository settings. MCP responses are redacted by default when serialized through the helper layer; callers should keep includeSecrets=false unless raw settings are explicitly needed.

Example CIFS planJson payload:

{
  "Name": "CIFS Share Crawl",
  "RepositoryType": "CIFS",
  "RepositorySettings": {
    "RepositoryType": "CIFS",
    "CifsHostname": "fileserver.example.com",
    "CifsUsername": "crawler",
    "CifsPassword": "secret",
    "CifsShareName": "content",
    "IncludeSubdirectories": true
  }
}

Example NFS planJson payload:

{
  "Name": "NFS Export Crawl",
  "RepositoryType": "NFS",
  "RepositorySettings": {
    "RepositoryType": "NFS",
    "NfsHostname": "nfs.example.com",
    "NfsUserId": 1000,
    "NfsGroupId": 1000,
    "NfsShareName": "/exports/content",
    "NfsVersion": "V3",
    "IncludeSubdirectories": true
  }
}

Route Coverage Matrix

Mapped means there is an MCP tool for the route family. Deferred means the route exists in REST but is intentionally not exposed from the current MCP release.

REST surface	MCP tools	Status	Notes
`GET /`, `HEAD /`, `GET /openapi.json`, `GET /v1.0/openapi.json`, `GET /v1.0/whoami`	`system/health`, `system/openapi`, `system/whoami`	Mapped	Health/head collapse into `system/health`; both OpenAPI routes return the same document
`GET /swagger`	None	Deferred	Browser Swagger UI; use `system/openapi` for the OpenAPI JSON from MCP
`POST /v1.0/authenticate`	`auth/authenticate`	Mapped	Useful for diagnosing upstream auth
`tenants` CRUD + HEAD	`tenant/list`, `tenant/get`, `tenant/create`, `tenant/update`, `tenant/delete`, `tenant/exists`	Mapped
tenant-scoped `users` CRUD + HEAD	`user/list`, `user/get`, `user/create`, `user/update`, `user/delete`, `user/exists`	Mapped	REST path is `/v1.0/tenants/{tenantId}/users...`
tenant-scoped `credentials` CRUD + HEAD	`credential/list`, `credential/get`, `credential/create`, `credential/update`, `credential/delete`, `credential/exists`	Mapped	REST path is `/v1.0/tenants/{tenantId}/credentials...`; `includeSecrets` opt-in
`buckets` CRUD + HEAD	`bucket/list`, `bucket/get`, `bucket/create`, `bucket/delete`, `bucket/exists`	Mapped
`bucket objects` list/put/delete/metadata/download/upload	`bucket/object/put`, `bucket/object/list`, `bucket/object/metadata`, `bucket/object/delete`, `bucket/object/download`, `bucket/object/upload`	Mapped	Binary transfers use base64
`collections` CRUD + HEAD + distinct metadata	`collection/list`, `collection/get`, `collection/create`, `collection/update`, `collection/delete`, `collection/exists`, `collection/labels/distinct`, `collection/tags/distinct`	Mapped
`collection records` list/get/create/delete/batch-delete	`collection/record/list`, `collection/record/get`, `collection/record/create`, `collection/record/delete`, `collection/record/batch-delete`	Mapped
`collection search`	`collection/search`	Mapped	Marshals RecallDB search requests through AssistantHub
`indices` CRUD + HEAD + labels/tags/custom metadata + top terms	`index/list`, `index/get`, `index/create`, `index/update`, `index/delete`, `index/exists`, `index/labels/update`, `index/tags/update`, `index/custom-metadata/update`, `index/terms/top`	Mapped	Marshals Verbex index requests through AssistantHub
`index records` list/get/create/batch-create/delete/batch-delete/HEAD/metadata	`index/record/list`, `index/record/get`, `index/record/create`, `index/record/create-batch`, `index/record/delete`, `index/record/batch-delete`, `index/record/exists`, `index/record/exists-batch`, `index/record/labels/update`, `index/record/tags/update`, `index/record/custom-metadata/update`	Mapped	AssistantHub uses `records`; Verbex upstream uses `documents`
`index search`	`index/search`	Mapped	Marshals Verbex search requests through AssistantHub
`assistants` CRUD + HEAD	`assistant/list`, `assistant/get`, `assistant/create`, `assistant/update`, `assistant/delete`, `assistant/exists`	Mapped
`assistant settings` get/update/slack verify + tool policy helpers	`assistant/settings/get`, `assistant/settings/update`, `assistant/settings/slack/verify`, `assistant/settings/tools/list`, `assistant/settings/tools/validate`, `assistant/settings/tools/test`	Mapped	`includeSecrets` opt-in; tool validation and dry-run diagnostics return redacted policy results and stable `ErrorCodes` by default
`assistant analytics` overview/timeseries/stages/endpoints/slowest/feedback	`assistantanalytics/overview`, `assistantanalytics/timeseries`, `assistantanalytics/stages`, `assistantanalytics/endpoints`, `assistantanalytics/slowest`, `assistantanalytics/feedback`	Mapped	Uses `assistantId` plus optional `AssistantAnalyticsQuery` JSON
`assistant tool-call traces` list/get/delete/bulk delete	`assistant/tool-calls/list`, `assistant/tool-calls/get`, `assistant/tool-calls/delete`, `assistant/tool-calls/delete-bulk`	Mapped	Redacted trace records only; supports `EnumerationQuery` filters for trace, tool, success, denied, chat-history, request-history, and time fields
`assistant public info + public documents + labels/tags`	`assistant/public/get`, `assistant/documents/list`, `assistant/labels/distinct`, `assistant/tags/distinct`	Mapped	Public metadata only; document list supports `queryJson`, text `query`, and `contentType` filters
`documents` list/get/upload/delete/HEAD/log/download/bulk-delete/reindex	`document/list`, `document/get`, `document/upload`, `document/delete`, `document/exists`, `document/processing-log`, `document/download`, `document/bulk-delete`, `document/reindex`, `document/reindex-batch`	Mapped	Binary transfers use base64; reindex tools backfill Verbex
`ingestion-rules` CRUD + HEAD	`ingestionrule/list`, `ingestionrule/get`, `ingestionrule/create`, `ingestionrule/update`, `ingestionrule/delete`, `ingestionrule/exists`	Mapped
`feedback` list/get/delete	`feedback/list`, `feedback/get`, `feedback/delete`	Mapped
`history` list/get/delete	`history/list`, `history/get`, `history/delete`	Mapped
`threads` list/get/create/delete	`thread/list`, `thread/get`, `thread/create`, `thread/delete`	Mapped
`requesthistory` list/summary/get/detail/delete/bulk-delete	`requesthistory/list`, `requesthistory/summary`, `requesthistory/get`, `requesthistory/detail`, `requesthistory/delete`, `requesthistory/bulk-delete`	Mapped
`embedding endpoints` CRUD + HEAD + health + test	`embeddingendpoint/list`, `embeddingendpoint/get`, `embeddingendpoint/create`, `embeddingendpoint/update`, `embeddingendpoint/delete`, `embeddingendpoint/exists`, `embeddingendpoint/health`, `embeddingendpoint/test`	Mapped
`embedding endpoint load`	None	Deferred	Use REST `POST /v1.0/endpoints/embedding/{endpointId}/load`
`completion endpoints` CRUD + HEAD + health + test	`completionendpoint/list`, `completionendpoint/get`, `completionendpoint/create`, `completionendpoint/update`, `completionendpoint/delete`, `completionendpoint/exists`, `completionendpoint/health`, `completionendpoint/test`	Mapped
`completion endpoint load`	None	Deferred	Use REST `POST /v1.0/endpoints/completion/{endpointId}/load`
`models` list/pull/pull-status/delete	`model/list`, `model/pull`, `model/pull/status`, `model/delete`	Mapped
`crawlplans` CRUD + HEAD + start/stop/connectivity/enumerate	`crawlplan/list`, `crawlplan/get`, `crawlplan/create`, `crawlplan/update`, `crawlplan/delete`, `crawlplan/exists`, `crawlplan/start`, `crawlplan/stop`, `crawlplan/connectivity`, `crawlplan/enumerate`	Mapped
`crawl plan draft connectivity`	None	Deferred	Use REST `POST /v1.0/crawlplans/connectivity` to test unsaved repository settings
`crawl operations` list/get/delete/statistics/enumeration	`crawloperation/list`, `crawloperation/get`, `crawloperation/delete`, `crawloperation/statistics`, `crawloperation/enumeration`	Mapped
`eval facts` CRUD	`eval/fact/list`, `eval/fact/get`, `eval/fact/create`, `eval/fact/update`, `eval/fact/delete`	Mapped
`eval runs` create/list/get/delete/results	`eval/run/create`, `eval/run/list`, `eval/run/get`, `eval/run/delete`, `eval/run/results`	Mapped
`eval result` get + judge prompt	`eval/result/get`, `eval/judge-prompt/default`	Mapped
`eval stream`	None	Deferred	Use REST SSE endpoint
`configuration` get/update	`configuration/get`, `configuration/update`	Mapped	`configuration/get` redacts by default
Public assistant `chat/open`, `chat`, `generate`, `compact`, `feedback`, `documents/{id}/download`	None	Deferred	Use REST directly; public document listing is mapped above

Example Tool Calls

Get runtime OpenAPI from MCP:

{
  "tool": "system/openapi",
  "arguments": {}
}

Get redacted runtime configuration:

{
  "tool": "configuration/get",
  "arguments": {}
}

Create a credential and intentionally return the bearer token:

{
  "tool": "credential/create",
  "arguments": {
    "tenantId": "ten_123",
    "credentialJson": "{\"UserId\":\"usr_123\",\"Name\":\"Automation key\",\"Active\":true}",
    "includeSecrets": true
  }
}

Download a document through the MCP wrapper:

{
  "tool": "document/download",
  "arguments": {
    "documentId": "adoc_123"
  }
}

Reindex one completed document into Verbex:

{
  "tool": "document/reindex",
  "arguments": {
    "documentId": "adoc_123"
  }
}

Reindex a page of completed documents into Verbex:

{
  "tool": "document/reindex-batch",
  "arguments": {
    "requestJson": "{\"IncludeAlreadyIndexed\":false}",
    "queryJson": "{\"MaxResults\":50}"
  }
}

Search the default Verbex text index:

{
  "tool": "index/search",
  "arguments": {
    "indexId": "default",
    "requestJson": "{\"Query\":\"deployment reset\",\"MaxResults\":10,\"IncludeMatchedTerms\":true,\"IncludeTermDetails\":true,\"IncludeDocumentTermStats\":true}"
  }
}

Search a RecallDB collection:

{
  "tool": "collection/search",
  "arguments": {
    "collectionId": "default",
    "requestJson": "{\"Query\":\"deployment reset\",\"MaxResults\":10}"
  }
}

Testing

Focused MCP validation:

$env:ASSISTANTHUB_TEST_SUITES="mcp"
dotnet run --project src/Test.Automated/Test.Automated.csproj

Preserve spawned server and MCP artifacts for investigation:

$env:ASSISTANTHUB_TEST_KEEP_ARTIFACTS="1"

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

AssistantHub MCP API Reference

Transports

Quick Start

Configuration

Tool Naming Rules

Secret Handling

Binary And Streaming Rules

Tool Families

Crawl Plan Repository Types

Route Coverage Matrix

Example Tool Calls

Testing

FilesExpand file tree

MCP_API.md

Latest commit

History

MCP_API.md

File metadata and controls

AssistantHub MCP API Reference

Transports

Quick Start

Configuration

Tool Naming Rules

Secret Handling

Binary And Streaming Rules

Tool Families

Crawl Plan Repository Types

Route Coverage Matrix

Example Tool Calls

Testing