Venom MCP plugin and server
Connect an agent to shared context, skills, and tools through the Venom MCP server.
The Venom MCP server.
Venom runs a remote MCP server over Streamable HTTP. Connecting requires a client that supports remote HTTP servers and Venom’s OAuth flow. Through that one connection, your agent gets projects, tasks, artifacts, skills, and the MCP servers your team has added to Venom.
When a client connects, the server sends short instructions on how to work in Venom: read the project and task before acting, create a task for new build work, log decisions, outcomes, and next steps, keep the status true, and check list_mentions. Clients without Venom Full's hooks also get a one-line reminder in every Venom tool response.
The server URL is the same for every organization. You choose the organization when you authorize. It's also shown in the web app under Settings → Agent connection.
https://mcp.venomlabs.dev/Claude Code or Codex?
Install Venom Full instead of adding the server by hand. It installs the venom plugin, which configures the connection, and adds hooks with session reminders. See Claude Code, Codex, and the Quickstart for the Codex commands.
Add the server to your client.
This covers claude.ai, Claude Desktop, and the Claude mobile apps. A connector you add on one appears on the others.
- In Claude, open Settings → Connectors and select Add custom connector.
- Name it
Venomand enterhttps://mcp.venomlabs.dev/as the URL. Leave the advanced OAuth settings empty. - Select Add, then Connect.
- Sign in to Venom in the window that opens, pick your organization, and approve.
On Team and Enterprise plans, an Owner first adds Venom under Organization settings → Connectors. Each member then connects their own Venom account from Settings → Connectors.
In a conversation, turn Venom on from the tools menu. Claude asks before running a tool that changes data. Tools that only read can run without asking. You can change both in the connector's settings.
Authorize.
The first time your client calls Venom, it starts an OAuth 2.1 flow:
- Your client registers itself with Venom and opens a browser window.
- You sign in with your Venom account.
- You choose the organization this connection works in, and approve.
- Your client receives a token and starts calling tools.
You don't copy or paste any keys. Your client refreshes the token on its own.
Things to know:
- One organization per connection. To work in another organization, authorize the connection again and pick it.
- The agent acts as you. It has your permissions in that organization, checked live on every call. If your membership is removed, the connection stops working.
- Agents are named in the log. Entries your agent writes show you as the author and the agent's name alongside.
Tools.
Tool names below are the ones Venom registers. Your client may add a prefix, such as mcp__venom__.
Find context.
| Tool | What it does |
|---|---|
list_teams | Lists your teams, or all teams in the organization when asked. |
list_projects | Lists projects from your teams and collaborations. Widens to a team or the organization only on request. |
list_tasks | Lists unfinished tasks in a project with their latest update, or all tasks. |
search | Finds literal, case-sensitive text in one project or task. |
Projects.
| Tool | What it does |
|---|---|
create_project | Creates a project owned by a team, with a key and an Overview. |
read_project | Reads the Overview, Log, task preview, and skill preview. Accepts the project key. |
update_project | Changes the Overview, name, or description, or archives and reactivates the project. |
append_project_log | Adds a project update, optionally attaching artifact revisions. |
Tasks.
| Tool | What it does |
|---|---|
create_task | Creates an open task with its brief. |
read_task | Reads the brief and recent entries. Accepts a task ref such as web-app-3. |
append_task_log | Records progress, decisions, blockers, or outcomes. Can change the status and replace the brief in the same step. |
edit_log_entry | Corrects, deletes, or restores an entry within its 30-day recovery window. |
Artifacts.
| Tool | What it does |
|---|---|
save_artifact | Saves a .md or .html deliverable as a new revision, with its log entry. |
list_artifacts | Finds artifacts on a project or task. |
read_artifact | Reads the latest or an exact revision. |
list_artifact_revisions | Lists revision history. |
set_artifact_archived | Archives or unarchives an artifact. |
Skills.
| Tool | What it does |
|---|---|
list_skills | Lists the skills for general work, a team, or a project. |
read_skill | Loads a skill's SKILL.md and, on request, its supporting files. With draft: true, people who can edit the skill get its latest saved version. |
create_skill | Creates a skill for you, a team, a project, or the organization. Personal skills publish right away; organization skills are always drafts. |
edit_skill | Changes a skill's files, and can publish team and project skills. |
venom_skill_<name> | Loads one published organization skill, for clients without native skills support. For example, venom_skill_release_notes for a skill named release-notes. |
Guidance.
| Tool | What it does |
|---|---|
get_guidance | Returns Venom's guidance for a client lifecycle event, such as compaction or a new subagent. Venom Full's hooks call it; see Hooks. |
MCP servers.
| Tool | What it does |
|---|---|
list_mcp_servers | Lists the MCP servers Venom provides in your current context, and whether each is ready to use. |
use_project | Switches to one project's MCP servers. Call it with no project to go back to all your teams'. |
Organization.
| Tool | What it does |
|---|---|
members | Lists active members with their handles and roles. |
organization_directory | Reads the full directory: profile, people, teams, and memberships. Owners and admins. |
update_organization_profile | Updates the organization name and description. Owners and admins. |
invite_members | Invites up to 10 people by email as members; Venom emails each a link. Leaves people already in the organization unchanged. Owners and admins. |
review_directory_batch | Validates and shows a batch of people and team changes for approval. Owners and admins. |
apply_directory_batch | Applies an approved batch, for clients that can't show the review screen. Owners and admins. |
Safe writes.
Memory writes, such as task updates and artifact saves, require a unique request_id. Retry an uncertain result with the same ID and identical arguments; changed arguments need a new ID.
Changes to existing memory also require the owner’s expected_version from a read. On a conflict, reread and reconcile before retrying. Skill edits use their own expected_version; skill tools do not take request_id. Follow each tool’s schema.
Your team's MCP servers.
Teams add other MCP servers, such as GitHub, Linear, or Sentry, on the MCP servers page in the web app. Like skills, a server lives in a place: yours, a team's, a project's, or the organization's.
Your agent doesn't connect to those servers directly. Venom passes their tools through the same connection, named <namespace>__<tool>, for example github__get_file_contents. Keys are stored encrypted in Venom and never reach your agent.
- With no project selected, your agent gets the organization's servers, your own, and all your teams'.
- After
use_project, it gets the organization's servers, your own, and that project's and its team's. - If a server needs your own sign-in or key, its tools stay hidden until you connect it on the MCP servers page.
list_mcp_serverstells the agent what's missing.