Skip to content
MCP server

Yodmcp

Python MCP server kernel adding persistent memory, tasks, skills and A2A to AI agents

Pricing
Open source
Platforms
CLI, API, self-hosted
Version
0.5.0
Listing updated
today

Similar MCP servers

About Yodmcp

Yodmcp is an open-source MCP server for agent tooling. It is available on CLI, API, self-hosted, Linux, macOS and Windows.

The README describes YodMCP as an agent operating system kernel that runs over MCP. Its core pieces are a multi-graph memory store with an optional SQLite-backed durable mode, a task handle system, and Agent Skills loaded from SKILL.md files and exposed as MCP resources.

The project ships several surfaces from one Python package: an MCP server over stdio or Streamable HTTP, a REST API on port 8080, an A2A agent card service, and a small Python SDK. Memory writes are gated so that promotion into semantic summaries depends on importance thresholds, and an audit trail and attestation records track tool activity. Attestation runs in software or simulated TEE modes by default, and the README states that Nitro and SGX are hooks that fall back to simulation without real providers.

The README is aimed at developers who want to attach a stateful kernel to Cursor, Claude Desktop or similar MCP clients, and who are prepared to run and verify the Python test suite themselves. It also documents soft quota plans and Stripe or USDC payment options for a hosted tier.

Tools this server provides (21)

Taken from the project documentation; the running server may differ.

memory_write
Insert one memory node for facts or decisions.
memory_read
Retrieve memory nodes by similarity or item id.
memory_delete
Hard-delete one memory node and its incident edges.
memory_consolidate
Promote high-importance nodes into semantic summaries.
memory_stats
Return node, edge and entity counts for memory.
tasks_create
Create a durable pending task handle.
tasks_get
Fetch one task handle by id.
tasks_list
List recent task handles with optional status filter.
tasks_update
Patch the status, progress or result of a task.
tasks_cancel
Mark a task handle as cancelled.
tasks_stats
Return task counts by status.
skills_list
List Agent Skills and their skills:// URIs.
a2a_card
Return the A2A Agent Card JSON.
plan_cache_get
Retrieve a semantic plan template.
plan_cache_put
Store a semantic plan template.
plan_cache_delete
Delete a semantic plan template.
cache_stats
Return in-process cache entry and hit totals.
attestation_recent
Return recent TRACE-style provenance claims.
audit_recent
Return recent policy and tool decision events.
discover_capabilities
List MCP tools and loaded skills.
echo
Liveness probe that echoes a message and version.

What it does

Multi-graph memory store
Stores facts and decisions as memory nodes with similarity retrieval, consolidation and an in-memory or SQLite backend.
Durable task handles
Creates, updates, lists and cancels task records that persist across sessions when the durable backend is used.
Agent Skills as resources
Loads SKILL.md files from a skills directory and exposes each skill body as a skills:// MCP resource.
Multiple server surfaces
Runs as an MCP server over stdio or HTTP, a REST API, an A2A agent card service and a Python SDK.
Audit and attestation records
Logs policy and tool decision events and produces provenance claims through a software or simulated TEE attestation mode.
OpenTelemetry instrumentation
Includes OpenTelemetry observability as listed in the project description.

Good for

  • Give a coding assistant in Cursor or Claude Desktop persistent memory across conversations
  • Track multi-step agent work as durable task handles with status and progress
  • Share reusable Agent Skills across MCP clients from a single skills directory
  • Expose an agent card and task endpoints for agent-to-agent messaging over A2A
  • Run a local verification suite to test memory, task and attestation behaviour before deployment

Pricing

Open source, from $49 a month

Apache-2.0 licensed and self-hosted; Pro at $49/mo and Enterprise at $499/mo are listed for a hosted plan.

Yodmcp FAQ

How do I connect YodMCP to Cursor?

The README gives a JSON mcpServers entry that runs the yodmcp command over stdio, with optional environment variables for the memory backend, plan and attestation mode. If the command is not on PATH, it shows a python -m yodmcp form with PYTHONPATH set to src.

Is the attestation real TEE protection?

No. The README says the nitro and sgx modes are provider hooks that fall back to simulated claims unless real TEE libraries are wired in, and it recommends software or simulated_tee for local verification.

Does memory persist between processes?

Not with the default in-memory backend. The README says processes do not share in-memory state and recommends the sqlite backend with a shared volume for multi-process durability.

What does the free plan include?

The README lists the free plan at zero dollars per month with 500 tool calls per day, and states that quotas are soft and tracked in-process.

Is Yodmcp free?

Yes. Yodmcp is open source, so you can use and self-host it for free.

What are the best alternatives to Yodmcp?

The closest alternatives to Yodmcp are mcp-memory-rs, Agent Memory MCP, mcp-gateway-registry, Superagenticmcp and Seamless.

Where this comes from

  • Description: written from the project's README and its MCP Registry entry.

How AIPoweredBot checks servers