generated: '2026-08-02'
method: searched
status: published
source: https://anaconda.com/docs/cli-reference/anaconda-mcp/getting-started, https://anaconda.com/docs/cli-reference/anaconda-ai/anaconda-ai-mcp,
  https://www.anaconda.com/blog/anaconda-mcp
notes: Anaconda ships two first-party MCP servers, both run locally. There is no hosted
  or remote MCP endpoint, so `tools/list` could not be introspected over HTTP; the
  tool set below is what Anaconda publishes in its own documentation.
servers:
- name: anaconda-mcp
  title: Anaconda MCP
  kind: gateway
  transport: stdio
  status: published
  distribution:
    registry: conda
    channel: anaconda
    package: anaconda-mcp
    install: conda create --name anaconda-mcp anaconda-mcp && conda activate anaconda-mcp
  setup: anaconda mcp setup --client claude-desktop
  clients_supported:
  - Claude Desktop
  - Claude Code
  - Cursor
  - VS Code
  - OpenCode
  - Windsurf
  cli_commands:
  - anaconda mcp setup
  - anaconda mcp clients
  - anaconda mcp remove
  docs: https://anaconda.com/docs/cli-reference/anaconda-mcp/getting-started
  description: A unified MCP gateway that composes multiple underlying conda-aware
    MCP servers into a single endpoint, giving an AI assistant access to the local
    conda environment state, `.condarc` channel configuration, installed packages,
    structured package metadata for real dependency resolution, PyPI-to-conda name
    mapping, and environment creation/cleanup.
  capabilities:
  - Query conda environments and enumerate installed packages
  - Create environments using the correct conda packages from the configured channels
    with proper dependency resolution
  - Respect the user's .condarc channel policy
  - Map PyPI package names to conda package names
  - Enumerate and clean up environments
  x-note: Anaconda's published documentation describes these capabilities in prose
    and does not enumerate verbatim MCP tool names for the gateway; no tool names are
    asserted here rather than inventing them. Authenticated `tools/list` introspection
    against a locally running gateway would resolve them.
- name: anaconda-ai-mcp
  title: Anaconda AI MCP
  kind: product
  transport: stdio
  alternate_transport: streamable-http
  default_http_host: 127.0.0.1
  default_http_port: 8000
  status: published
  distribution:
    registry: pypi
    package: anaconda-ai
    extra: mcp
    install: pip install 'anaconda-ai[mcp]'
  command: anaconda ai mcp
  example: anaconda ai mcp --transport streamable-http --port 9000
  docs: https://anaconda.com/docs/cli-reference/anaconda-ai/anaconda-ai-mcp
  backing_api: anaconda:ai-navigator-api
  description: Exposes the Anaconda AI Navigator model catalog and local inference-server
    lifecycle as MCP tools.
  tools:
  - name: list_models
    description: List the curated model catalog / locally downloaded models
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#GET /api/models
  - name: download_model
    description: Download a model file to the local machine
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#PATCH /api/models/{modelId}/files/{fileId}
    source_confidence: medium
  - name: list_servers
    description: List local inference servers
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#GET /api/servers
  - name: start_server
    description: Start a local inference server
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#PATCH /api/servers/{serverId}
  - name: stop_server
    description: Stop a running local inference server
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#PATCH /api/servers/{serverId}
  - name: remove_server
    description: Delete a local inference server
    source_operation: openapi/anaconda-ai-navigator-openapi-original.json#DELETE /api/servers/{serverId}
  x-note: 'Tool names are quoted verbatim from Anaconda''s command reference: "MCP
    tool names use underscores (Python convention) and may differ from CLI command
    names."'
related:
- name: Agent Studio MCP servers
  docs: https://anaconda.com/docs/anaconda-desktop/agent-studio/mcp-servers
  note: Anaconda Desktop's Agent Studio is also an MCP *client* — it can attach third-party
    MCP tool servers to its agents.
x-evidence:
  fetched: '2026-08-02'
  hosted_endpoint: none published
  tools_list_introspected: false
  reason: both servers are local (stdio by default); no remote URL to POST tools/list
    against