vector-mcp

Action-routed MCP and agent interfaces for governed vector collection management and retrieval. The native default is epistemic-graph. Secure opt-in providers cover PostgreSQL/pgvector, Qdrant, and MongoDB Atlas.

Version: 3.0.0

Governed capability

• MCP tools: vector_collection_management and vector_search

• Skill provider: the consolidated vector-mcp-operations workflow

• Ontology provider: the packaged vector retrieval ontology

• Source connector provider: a read-only vector collection inventory preset

• Runtime configuration: AgentConfig, environment variables, and secret references

• Privacy posture: no checked-in endpoints, credentials, personal identity, or host paths

Related MCP server: production-grade-mcp-agentic-system

Install

Use the smallest extra set required by the deployment:

uvx --from 'vector-mcp[mcp]' vector-mcp

The runtime requires agent-utilities>=2.0.0 and its self-contained full epistemic-graph engine contract. A bare numeric-only or partial engine profile is not a supported deployment.

For a selected storage provider:

uv add 'vector-mcp[postgres]'
uv add 'vector-mcp[qdrant]'
uv add 'vector-mcp[mongodb]'

The all extra enables every supported optional provider plus the agent, Langfuse, and Logfire runtimes. Production images should install only the providers they operate.

MCP configuration

The package includes a neutral agent-launch configuration containing only the command, condensed tool mode, and tool toggles. Runtime values are inherited from AgentConfig or injected by the operator. Detailed instructions on how to use the underlying API wrappers, extended schema bindings, and developer SDK references are maintained in docs/index.md.

MCP

This server utilizes dynamic Action-Routed tools to optimize token overhead and maximize IDE compatibility.

Available MCP Tools

Auto-generated from the live MCP server — do not edit by hand.

Condensed action-routed tools (default — MCP_TOOL_MODE=condensed)

Verbose 1:1 API-mapped tools (MCP_TOOL_MODE=verbose or both)

1 action-routed tool(s) (default) · 7 verbose 1:1 tool(s). Each is enabled unless its <DOMAIN>TOOL toggle is set false; MCP_TOOL_MODE selects the surface (condensed default · verbose 1:1 · both). Auto-generated — do not edit.

Detailed tool schemas, parameter shapes, and validation constraints are preserved in docs/mcp.md.

Dynamic Tool Selection & Visibility

This MCP server supports dynamic toolset selection and visibility filtering at runtime. This allows you to restrict the set of exposed tools in order to prevent blowing up the LLM's context window.

You can configure tool filtering via multiple input channels:

• CLI Arguments: Pass --tools or --toolsets (or their disabled counterparts --disabled-tools and --disabled-toolsets) during startup.

• Environment Variables: Define standard environment variables:

  • MCP_ENABLED_TOOLS / MCP_DISABLED_TOOLS

  • MCP_ENABLED_TAGS / MCP_DISABLED_TAGS

• HTTP SSE Request Headers: Pass custom headers during transport initialization:

  • x-mcp-enabled-tools / x-mcp-disabled-tools

  • x-mcp-enabled-tags / x-mcp-disabled-tags

• HTTP SSE Request Query Parameters: Append query parameters directly to your transport connection URL:

  • ?tools=tool1,tool2

  • ?tags=tag1

When query strings or parameters are supplied, an LLM-free Knowledge Graph resolution layer (using DynamicToolOrchestrator) matches query intents against known tool tags, names, or descriptions, with safe fallback and automated 24-hour background cache refreshing.

MCP Configuration Examples

Install the slim [mcp] extra. All examples install vector-mcp[mcp] — the MCP-server extra that pulls only the FastMCP / FastAPI tooling (agent-utilities[mcp]). It deliberately excludes the heavy agent runtime (pydantic-ai, the epistemic-graph engine, dspy, llama-index), so uvx / container installs are far smaller. Use the full [agent] extra only when you need the integrated Pydantic AI agent.

stdio Transport (local IDEs — Cursor, Claude Desktop, VS Code)

A client launch entry can remain equally small:

{
  "mcpServers": {
    "vector-mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "vector-mcp[mcp]",
        "vector-mcp"
      ],
      "env": {
        "MCP_TOOL_MODE": "condensed",
        "COLLECTION_MANAGEMENTTOOL": "True",
        "SEARCHTOOL": "True",
        "DATABASE_TYPE": "epistemic_graph"
      }
    }
  }
}

Streamable-HTTP Transport (networked / production)

{
  "mcpServers": {
    "vector-mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "vector-mcp[mcp]",
        "vector-mcp",
        "--transport",
        "streamable-http",
        "--port",
        "8000"
      ],
      "env": {
        "TRANSPORT": "streamable-http",
        "HOST": "0.0.0.0",
        "PORT": "8000",
        "MCP_TOOL_MODE": "condensed",
        "COLLECTION_MANAGEMENTTOOL": "True",
        "SEARCHTOOL": "True",
        "DATABASE_TYPE": "epistemic_graph"
      }
    }
  }
}

Alternatively, connect to a pre-deployed Streamable-HTTP instance by url. Do not put concrete endpoints, certificate paths, credentials, or user directories in the repository. Durable credentials and TLS profiles should be supplied by secret reference.

Tool surface

vector_collection_management

Supported actions:

• create_collection

• add_documents

• delete_collection

• list_collections

Database credentials and local database paths are not accepted as tool arguments. Document files are selected only by paths relative to the administrator-owned DOCUMENT_DIRECTORY. Absolute paths, URLs, traversal, symbolic links, unbounded file sets, and oversized content are rejected before delegation.

vector_search

Supported actions:

• semantic_search

• lexical_search

• search (hybrid retrieval)

Backend names, collection identifiers, result counts, search weights, and request sizes are validated at the MCP boundary. Legacy aliases and providers without the common indexed, authenticated, verified-TLS contract are not advertised or accepted.

Runtime trust

Embedding providers are selected through the shared AgentConfig EMBEDDING_MODELS registry. Model credentials remain references in that registry, and model trust is selected through EMBEDDING_TLS_PROFILE or EMBEDDING_TLS_PROFILE_REF. Database credentials likewise use only env://, vault://, or secret:// references. Complete-chain PEM bundles, system trust, mTLS, and proxy policy are resolved by Agent Utilities at runtime; boolean certificate-verification bypasses are not supported. The doctor reports configuration booleans and readiness without printing endpoints, hostnames, paths, identities, or secret references:

vector-mcp-doctor

Auto-generated from the code-read env surface (MCP_TOOL_MODE + package vars) — do not edit.

Additional Deployment Options

vector-mcp can also run as a local container (Docker / Podman / uv) or be consumed from a remote deployment. The Deployment guide has full, copy-paste mcp_config.json for all four transports — stdio, streamable-http, local container / uv, and remote URL:

• Local container / uv — launch the server from mcp_config.json via uvx, docker run, or podman run, or point at a local streamable-http container by url.

• Remote URL — connect to a server deployed behind Caddy at http://vector-mcp.arpa/mcp using the "url" key.

Environment Variables

Package environment variables

Inherited agent-utilities variables (apply to every connector)

27 package + 14 inherited variable(s). Auto-generated from .env.example + the shared agent-utilities set — do not edit.

Every variable the server reads, grouped by purpose. See .env.example for the canonical, copy-paste list — including the DATABASE_TYPE / GRAPH_SERVICE_SOCKET / GRAPH_SERVICE_AUTH_SECRET connection settings for the native epistemic-graph backend. Backend endpoints, database locations, and credentials for opt-in providers (Postgres/Qdrant/Mongo/ Chroma/Couchbase) are never README-documented literal values or MCP tool arguments — they resolve through AgentConfig and secret:///env:///vault:// references at runtime.

MCP server / transport

Tool toggles

Each action-routed tool can be disabled individually via its toggle env var (set to false). The full list is in the Available MCP Tools table above.

Telemetry & governance

Agent CLI (full [agent] runtime only)

See .env.example for a copy-paste starting point.

Provider and ontology integration

The package contributes its skills, prompts, ontology, and source connector through Python entry points. The collection-inventory connector is intentionally read-only and registers collection metadata, not document or embedding payloads.

Generated connector signatures must be recreated only after the installed MCP schema is observed and a release signing key is provided at runtime. A signature from an older tool schema or ontology must never be copied forward.

Development checks

Low-cost checks that do not launch providers:

python scripts/security_sanitizer.py
python scripts/security_contract.py --contract .security/security-contract.json validate
python -m compileall -q vector_mcp

Provider tests use mocked SDK boundaries and make no network calls. Live qualification is a separate deployment gate and must use operator-supplied AgentConfig and secrets.

Documentation

• Installation

• Configuration and privacy

• Deployment

• Usage

• Architecture overview

The slim :mcp streamable-http container (docker/mcp.compose.yml) publishes :8000 with a /health check; see Deployment for the full compose service definition.

License

See LICENSE.

Deploy with agent-utilities-deployment

Provision this package with the consolidated agent-utilities-deployment workflow. It selects an installed-package, editable-source, or immutable-container path; records only runtime secret and TLS-profile references in AgentConfig; and runs doctor, registration, policy, observability, and rollback gates. Ask your agent to "deploy vector-mcp with agent-utilities-deployment".

The repository embeds no deployment profile, credential value, certificate path, or environment-specific endpoint. Supply those at runtime through AgentConfig and the configured secret provider.

Installation

Pick the extra that matches what you want to run:

# MCP server only (recommended for tool hosting — slim deps)
uv pip install "vector-mcp[mcp]"

# Full agent runtime (Pydantic AI + epistemic-graph engine)
uv pip install "vector-mcp[agent]"

# Everything (development)
uv pip install "vector-mcp[all]"      # or: python -m pip install "vector-mcp[all]"

Container images (:mcp vs :agent)

One multi-stage docker/Dockerfile builds two right-sized images, selected by --target:

docker build --target mcp   -t knucklessg1/vector-mcp:mcp    docker/   # slim MCP server
docker build --target agent -t knucklessg1/vector-mcp:latest docker/   # full agent

docker/mcp.compose.yml runs the slim :mcp server; docker/agent.compose.yml runs the agent (:latest) with a co-located :mcp sidecar.

Knowledge-graph database (epistemic-graph)

The full agent ([agent] / :latest) embeds the epistemic-graph engine (pulled in transitively via agent-utilities[agent]). For production — or to share one knowledge graph across multiple agents — run epistemic-graph as its own database container and point the agent at it instead of embedding it. Deployment recipes (single-node + Raft HA), connection config, and the full database architecture (with diagrams) are documented in the epistemic-graph deployment guide. The slim [mcp] server does not require the database.

Repository Owners

Contribute

Contributions are welcome! Please ensure code quality by executing local checks before submitting pull requests:

• Format code using ruff format .

• Lint code using ruff check .

• Validate type-safety with mypy .

• Execute test suites using pytest

Deploy with agent-os-genesis

This package can be provisioned for you — skill-guided — by the agent-os-genesis universal skill (its single-package deploy mode): it picks your install method, seeds secrets to OpenBao/Vault (or .env), trusts your enterprise CA, registers the MCP server, and verifies it — the same machinery that stands up the whole Agent OS, narrowed to just this package. Ask your agent to "deploy vector-mcp with agent-os-genesis".

Secrets are read-existing + seeded via vault_sync — you are only prompted for what's missing.