M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
M

Managed Agents MCP Server

Community
A local-first AI agent runtime with an MCP bridge.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #315 of 438
Stars
★ 707 Very popular
Transport
stdio (local process)
Runtime
Node.js 22+ · Docker
Credentials
Optional API key
License
Apache-2.0
Last commit
44FMRS · D

The supplied sources describe a local agent runtime with MCP tools and a native DeepSeek Harness stdio bridge. No concrete MCP tool identifiers are listed, so the tool list is empty.

Strongest · Documentation 11/20 Weakest · Maintenance 7/20

Reliability
8/20
Security and permissions
8/20
Maintenance
7/20
Documentation
11/20
Setup experience
10/20
Why each score
Reliability 8/20
The README describes a stdio MCP bridge, tool discovery/use cases, and typecheck, test, build, and release-check commands, but provides no executed results, CI workflows, committed source, or MCP initialization/tool-enumeration test evidence. The manifest is unavailable, so declared tools cannot be compared with implementation; startup and handshake are therefore only plausible, with deductions for unverified error handling, dependency control, and edge cases.
Security and permissions 8/20
The README makes API keys, a credential vault, permission policies, sandbox backends, audit trails, and external MCP connections visible, so the main security capabilities and data-flow risks are disclosed. However, the runtime is open by default, authentication activates only when an API key exists, and there is no evidence for least-privilege defaults, confirmation of dangerous operations, network isolation, credential-protection implementation, or cross-session data boundaries. No explicit red-line risk is shown, but the missing evidence warrants substantial deductions.
Maintenance 7/20
The repository is not archived and is reported with 571 stars and 1 open issue; the README claims a CHANGELOG, an Apache-2.0 LICENSE, and a release:check. However, there is no evidence of commit or release cadence, dependency updates, a security-response channel, or maintainer/ownership details, and the repository metadata says NOASSERTION for the license, conflicting with the README. Only limited maintenance credit is justified.
Documentation 11/20
The README covers installation, operation, API usage, SDK access, authentication, configuration, environments, DeepSeek Harness integration, and links to additional documentation, with several examples. It does not provide detailed MCP tool parameters and error responses, limits, costs, troubleshooting, platform compatibility, or verifiable evidence of tool implementation; the repository description and README project naming also differ. The documentation is usable but incomplete.
Setup experience 10/20
The README gives Node.js/npm requirements, npx initialization and startup commands, a local endpoint, API examples, and a DSH plugin flow, making the main path reasonably clear. It still requires a model-provider key, optional Docker/Kubernetes backends add complexity, and there is no mainstream MCP client configuration example, platform matrix, or verified connection evidence. Under the static rule, the absence of CI and key-path test evidence prevents a high score.

Static review · not runListed 2026-08-14

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkConnects to a database

Best for

  • Developers needing a local or self-hosted agent runtime
  • Teams requiring persistent sessions, auditability, and replay
  • Users working with DeepSeek Harness and MCP

Not for

  • Users seeking a hosted model service
  • Users seeking a visual workflow builder
  • Environments without Node.js 22+ or a model-provider API key

Required permissions

  • Access to the local runtime and its workspace state directory
  • Depending on configuration, local-process, Docker, Kubernetes, or self-hosted worker permissions
  • A configured model-provider API key for model execution

Risks and side effects

  • Misconfigured sandbox backends may broaden tool access to the host or infrastructure
  • Credentials, files, memory, and audit data are stored in the runtime workspace or configured storage backends
  • After API-key authentication is enabled, key holders can make authorized runtime requests

Setup

Before you start

Runtime：Node.js 22+ · Docker

MANAGED_AGENTS_API_KEY optionalsecret API key for the managed-agents runtime; authentication activates when a key exists and clients send it as a Bearer token, or keys can be created via /v1/api-keys.
OPENAI_API_KEY optionalsecret Model provider API key (OpenAI, Anthropic, or OpenAI-compatible endpoint also works), configured in Settings > Models or config.yaml.
Other optional settings (1)
MANAGED_AGENTS_URL optional URL of the running managed-agents runtime (e.g. http://127.0.0.1:3000), used by the DSH plugin to connect.

Set MANAGED_AGENTS_URL=http://127.0.0.1:3000. Install the bundle into a DSH profile with dsh plugin --profile web add managed-agents, then run dsh web. Start managed-agents before using the profile.

Check that it works

After installing the DSH plugin and booting the profile, mcp__sandbase__* tools (e.g. listing agents) should appear in the client's tool list; you can also confirm the runtime is up by running managed-agents start and opening http://127.0.0.1:3000/dashboard.

Troubleshooting

  1. Verify that Node.js 22+ and npm 10+ are installed
  2. Verify that managed-agents is running and MANAGED_AGENTS_URL is correct
  3. Check the model provider and API key under Settings > Models
  4. If using Docker, verify Docker availability; if using Kubernetes, verify kubectl and cluster permissions
  5. Check the DSH profile installation and logs for the stdio MCP process

Things to try

Once connected, you can ask your AI assistant things like:

  • List the available agents
  • Create a session and have it investigate this alert
  • Show the results and artifacts from the last session
  • Stop the currently running task

Use cases

Manage and run AI agent sessions
Execute tools through sandbox backends
Inspect session results, events, and artifacts
Use DeepSeek Harness to call the runtime

Supported clients

DeepSeek Harness

Listed from the project's documentation, not tested by this site.

Overview

Managed Agents is a local-first AI agent runtime providing persistent sessions, sandboxed tools, memory, credentials, audit trails, replay, and a local Console, with DeepSeek Harness integration over stdio MCP.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 49d73de015b3 Data synced 2026-10-12 Read the FMRS scoring method