Flagsmith MCP server: 37 tools, 36 undescribed

A generated server the vendor never finished publishing, what its empty tool descriptions do to an agent, and what to use instead.

Last verified 1 September 2026 · from Flagsmith's repository and the server's own tools/list response, built from source

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Flagsmith MCP server?

It is a server for the Model Context Protocol. Flagsmith is a feature-flag platform, so a working connector would let an agent read and change flags without opening the dashboard.

This one was generated rather than written. Every tool maps to one Flagsmith REST endpoint, and the names are the endpoint paths with slashes replaced: mcp-api-v1-projects-features-list, mcp-api-v1-projects-segments-create, and so on.

Two endpoints can produce the same tool name. The generator then appends a hash to keep them apart. That is why mcp-api-v1-environments-features-versions-3b3, -e40 and -34b sit next to each other with nothing to tell them apart.

Flagsmith is a release tool, so it sits in CI and release, and reaches nothing in code review, production observability, agent observability or cost.

For an agent, the tools reach everything and explain nothing. The endpoints behind these 37 tools are the same ones the Flagsmith API exposes, and the API documents them. The server strips that documentation out, so the agent gets the reach without the instructions.

What can an AI agent do with Flagsmith?

Every row above is reachable and none is guided. The Flagsmith REST API documents these endpoints properly. An agent with the API and a fetch tool is better informed than one with this server.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

37 tools cost between 11,213 and 20,671 tokens, which is 5.6% to 10.3% of a 200,000-token window.

All of that is schema. Descriptions average 2 characters across the 37, because only one tool has any description at all; input schemas average 1,073.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted 24 August 2026 from tools/list on the server built from the repository at its 13 March 2026 state. The server ran locally with a placeholder token.

Registration is static, so no Flagsmith account was involved. Figures cover name, description and input schema. The two models in column one share the o200k_base encoding, so they get one count. The spread is formatting, about 63%. Reasoning tokens are output, spent per request. The tool list is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer, so the definitions were sent to the API's count_tokens endpoint and a no-tools baseline subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end, Opus 5 and Sonnet 5 at the high end. The gap between them is about a third.

Flagsmith MCP server limits: where answers come back incomplete

Every limit below comes from a generated server that was never edited after generation.

How to configure Flagsmith MCP for agents

1. 1

   Decide whether to run an unpublished server at all

   There is no release, no tag and no package, so what you run is whatever the default branch contained on the day you cloned it.

2. 2

   Expect no guidance from the tools

   36 of 37 descriptions are empty, so any reliability has to come from your own prompt naming the tools you want used.

3. 3

   Treat the writing tools carefully

   11 names contain create, update or destroy, and none of them explains what it changes.

4. 4

   Consider the REST API instead

   Flagsmith documents the same endpoints properly, which is more than this server passes on.

Paste this into your agent’s instructions

With no descriptions and hashed names, an agent cannot explain why it picked a tool. The block below makes it name the tool it called.

When you answer using Flagsmith MCP tools, state:
- The exact tool name you called. Several differ only by a
  three-character hash at the end.
- The project or environment the answer covers.
- Whether the tool changed a flag, segment or project. Names
  containing create, update or destroy write.
Never state that data does not exist. State that the tool
above returned no results.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Yes, and here the alternative is better documented than the connector. Flagsmith's REST API describes the same endpoints with parameters and examples, which is exactly what this server strips out.

An agent with a fetch tool and the API reference can do everything these 37 tools do, and can read what each endpoint expects.

If you want a flag connector that is written rather than generated, LaunchDarkly publishes one with descriptions on every tool.

Use the Flagsmith REST API. Reach for this server only if you intend to maintain your own build of it.

Flagsmith MCP server setup

There is nothing to install from a registry. Clone the repository, install its dependencies and build it, then run the built entry point.

The server takes a subcommand: start --transport stdio for a local client. Running the binary with no subcommand prints usage and exits.

claude mcp add-json "flagsmith" '{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js",
          "start","--transport","stdio",
          "--token-auth","<your-token>"]}'

{"mcpServers":{"flagsmith":{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js",
          "start","--transport","stdio",
          "--token-auth","<your-token>"]}}}

[mcp_servers.flagsmith]
command = "node"
args = ["/path/to/flagsmith-mcp/bin/mcp-server.js",
        "start", "--transport", "stdio",
        "--token-auth", "<your-token>"]

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"flagsmith":{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js","start","--transport","stdio","--token-auth","<your-token>"]}}}

{"mcpServers":{"flagsmith":{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js","start","--transport","stdio","--token-auth","<your-token>"]}}}

{"servers":{"flagsmith":{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js","start","--transport","stdio","--token-auth","<your-token>"]}}}

{"mcpServers":{"flagsmith":{"command":"node",
  "args":["/path/to/flagsmith-mcp/bin/mcp-server.js","start","--transport","stdio","--token-auth","<your-token>"]}}}

git clone https://github.com/Flagsmith/flagsmith-mcp.git
cd flagsmith-mcp && npm install && npm run build
# produces bin/mcp-server.js

node bin/mcp-server.js start --transport stdio \
  --token-auth <your-token> \
  --tool mcp-api-v1-projects-features-list

Read-only access, permissions and security

• You are running an unreviewed build

  No release, no tag and no published package means no fixed version to audit or pin.

• 11 tools write

  Names containing create, update or destroy change flags, segments, projects and change requests, and none describes what it does.

• Empty descriptions make approval harder

  A client that asks you to approve a tool call can only show you a name such as mcp-api-v1-environments-features-versions-e40.

• The README badge and the repository disagree

  The README shows an MIT badge while GitHub reports no license for the repository.

Troubleshooting

The server prints usage and exits

It needs a subcommand. Use start --transport stdio for a local client, or serve for HTTP.

There is no package to install

That is correct. The repository has 0 releases and 0 tags and nothing on npm, so cloning and building is the only route.

The agent calls the wrong tool repeatedly

36 of 37 tools have no description, and eight names differ only by a trailing hash. Register a narrow set with repeated --tool flags.

A flag changed unexpectedly

11 tool names contain a writing verb. The Flagsmith token's permissions are the only boundary the server offers.

Flagsmith MCP server: Reference

What engineers report

There is no adoption to report, which is itself the finding: 0 stars, 0 releases and no package.

Repository facts from the GitHub API, retrieved 24 August 2026, including a releases count of 0 and a tags count of 0. The npm registry was checked for flagsmith-mcp and @flagsmith/mcp on the same date and returned 404 for both. Tool counts and description lengths come from the server's own tools/list response on a build of the repository.

Should you connect an agent to Flagsmith?

Yes, though there is no packaged release. Best for not connecting. The documented REST API gives an agent more to work with than this server does.

Strongest fit

Teams willing to maintain their own build, who will name every tool in the prompt themselves.

Main advantage

It does reach the whole Flagsmith API surface, because it was generated from the specification.

Main weakness

36 of 37 tools carry no description, so the model has nothing to choose from but a REST path.

Operational risk

Eleven undescribed writing tools, on an unreleased build with no version to pin.

Frequently asked questions